GitLab

对称参照 Claude BetaPolicy 的 fast-mode 过滤实现，新增针对 OpenAI 上游
service_tier 字段（priority / flex，含客户端 "fast" → "priority" 归一化）的
pass / filter / block 三态策略，覆盖全部 OpenAI 入口 + admin 配置入口。

后端核心
- 新增 SettingKeyOpenAIFastPolicySettings、OpenAIFastPolicyRule、
  OpenAIFastPolicySettings 配置模型，含规则的 service_tier × action × scope
  × 模型白名单 × fallback action 维度。
- SettingService.Get/SetOpenAIFastPolicySettings；缺失时返回内置默认策略
  （所有模型的 priority 走 filter，whitelist 为空，fallback=pass）。设计
  依据：service_tier=fast 是用户级开关，与 model 字段正交，默认锁定特定
  model slug 会留下"用 gpt-4 + fast 透传 priority 上游"的绕过路径。JSON
  解析失败不再静默 fallback，slog.Warn 记录脏数据，便于运维定位。
- service_tier 归一化（trim + ToLower + fast→priority + 白名单 priority/flex）
  与策略评估（evaluateOpenAIFastPolicy）作为唯一真实来源，HTTP / WS 共用。
  抽出纯函数 evaluateOpenAIFastPolicyWithSettings，配合 ctx-bound settings
  快照（withOpenAIFastPolicyContext / openAIFastPolicySettingsFromContext），
  WS 长会话入口预取一次后所有帧复用，避免每帧打到 settingService。

HTTP 入口（4 个）
- Chat Completions、Anthropic 兼容（Messages，含 BetaFastMode→priority 二次
  命中）、原生 Responses、Passthrough Responses 全部接入
  applyOpenAIFastPolicyToBody，filter 走 sjson 顶层删除 service_tier，block
  返回 403 forbidden_error JSON。
- 4 入口统一使用 upstream 视角的 model（GetMappedModel +
  normalizeOpenAIModelForUpstream + Codex OAuth normalize 后的 slug），
  避免 chat/messages/native /responses/passthrough 因为 model 维度不同
  造成 whitelist 命中差异。
- 在 pass 路径也把客户端 "fast" 别名归一化为 "priority" 写回 body，
  否则 native /responses 与 passthrough 入口会把 "fast" 原样透传给上游
  导致 400/拒绝（chat-completions 入口的 normalizeResponsesBodyServiceTier
  此前已具备同等行为）。

WebSocket 入口
- 新增 applyOpenAIFastPolicyToWSResponseCreate：严格匹配
  type="response.create"，仅处理顶层 service_tier；filter 用 sjson 删字段，
  block 返回 typed *OpenAIFastBlockedError。
- ingress 路径在 parseClientPayload 内调用，block 命中先 Write Realtime
  风格 error event 再返回 OpenAIWSClientCloseError(StatusPolicyViolation
  =1008)，依赖底层 WebSocket Conn.Write 的同步 flush 保证 error 先于
  close。
- passthrough 路径在 RunEntry 前对 firstClientMessage 应用策略，并通过
  openAIWSPolicyEnforcingFrameConn 包装 ReadFrame 对每个 client→upstream
  帧执行策略；后续帧无 model 字段时回退到 capturedSessionModel。
  filter 闭包内同时侦测 session.update / session.created 帧的 session.model
  字段刷新 capturedSessionModel，封堵"首帧 model=gpt-4o（pass）→
  session.update 改为 gpt-5.5 → 不带 model 的 response.create fallback
  到 gpt-4o"的 mid-session 绕过路径。
- passthrough billing：requestServiceTier 在策略 filter 之后再从
  firstClientMessage 提取，filter 命中时 OpenAIForwardResult.ServiceTier
  上报 nil（default tier），与 HTTP 入口（reqBody 来自 post-filter map）
  / WS ingress（payload 来自 post-filter bytes）的语义一致。
- 错误事件 schema：{event_id: "evt_<32hex>", type: "error",
  error: {type: "forbidden_error", code: "policy_violation", message}}，
  与 OpenAI codex 客户端 error event 解析兼容。

Admin / Frontend
- dto.SystemSettings / UpdateSettingsRequest 新增
  openai_fast_policy_settings 字段（omitempty），bulk GET/PUT 接入。
- Settings 页 Gateway 页签新增 Fast/Flex Policy 表单卡片：
  service_tier × action × scope × 模型白名单 × fallback action 全字段配置。
- 前端守门：openaiFastPolicyLoaded 标志仅在 GET 真带回字段时才允许回写，
  避免 rollout/错误把默认规则覆盖成空；saveSettings 回写循环 skip 该字段，
  由专用刷新逻辑处理；仅 action=block 时发送 error_message，匹配后端
  omitempty 行为。

测试
- HTTP 路径：openai_fast_policy_test.go 覆盖默认配置（whitelist=[]，所有
  模型 priority filter）/ block 自定义错误 / scope 区分 / filter 删字段 /
  block 不改 body / block 短路上游 / Anthropic BetaFastMode 触发 OpenAI
  fast policy 等场景。
- WebSocket 路径：openai_fast_policy_ws_test.go 覆盖
    helper 单元（filter / fast→priority 归一化 / flex 透传 / block typed
    error / 无 service_tier 字节不变 / 非 response.create 帧不动 / 空 type
    帧不动 / event_id+code 字段断言 / 非字符串 service_tier 容错）+
    pass 路径 fast 别名归一化回归 +
    ingress 端到端（filter 后上游不含 service_tier / block 后客户端先收
    error event 再收 close 1008 且上游 0 写）+
    passthrough capturedSessionModel fallback 用例（whitelist 策略下首帧
    建立、缺 model 命中 fallback、缺少 fallback 时的 leak 文档化）+
    passthrough session.update / session.created 旋转 capturedSessionModel
    的 mid-session 绕过回归 +
    passthrough billing post-filter ServiceTier 与 idempotent filter 回归。
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

- 在系统设置「功能开关」中新增邀请返利总开关，默认关闭；
  关闭态：菜单隐藏、注册忽略 aff、新充值不返利，但已有 quota 仍可转余额
- 支持管理员为指定用户设置专属邀请码（覆盖随机码，全局唯一）
- 支持管理员为指定用户设置专属返利比例（覆盖全局比例，可单条/批量调整）
- 在系统设置邀请返利卡片内嵌入专属用户管理表格（搜索/编辑/批量/删除），
  删除采用项目通用 ConfirmDialog，会同时清除专属比例并把邀请码重置为系统随机码
- /affiliate 用户页新增「我的返利比例」卡片与动态使用说明，让用户直观看到
  分享后能拿到多少（同源 resolveRebateRatePercent 计算，与实际充值一致）
- 新增数据库迁移 132 添加 aff_rebate_rate_percent 与 aff_code_custom 列
- 新增 admin 路由组 /api/v1/admin/affiliates/users/* 共 5 个端点
- AffiliateService 改为只依赖 *SettingService，去除冗余的 SettingRepository
- 邀请码格式校验放宽到 [A-Z0-9_-]{4,32}，兼容旧 12 位系统码与新自定义码
- 补充单元测试与集成测试覆盖新方法、冲突路径与边界值

- Drop SetAffiliateService setters and ProvideAuthService /
  ProvidePaymentService / ProvideUserHandler wrappers in favor of direct
  Wire constructor injection. AffiliateService has no back-edge to
  Auth/Payment/User, so the indirection was never required.
- Change RegisterWithVerification's variadic affiliateCode to a fixed
  parameter; adjust all call sites.
- Validate aff_code length and charset in BindInviterByCode before any
  DB lookup, eliminating timing-side-channel and useless DB roundtrips
  on malformed input.
- Make affiliate cache invalidation synchronous; surface Redis errors
  via the project logger instead of swallowing them in a detached
  goroutine.
- Add an integration test guarding cross-layer tx propagation in
  AccrueQuota and a unit test pinning the aff_code format rules.

- Extract PublicSettingsInjectionPayload named struct with drift test
- Add channel_monitor_default_interval_seconds to SSR injection
- Add image_output_price to SupportedModelChip
- Simplify AppSidebar buildSelfNavItems (admins see available channels)
- Add gateway WARN logs for 503 no-available-accounts branches
- Wire ChannelMonitorRunner into provideCleanup for graceful shutdown
- Add migrations 130/131 (CC template userid fix + mimicry field cleanup)
- Clean up fork-only features (sora, claude max simulation, client affinity)
- Remove ~320 obsolete i18n keys
- Add codexUsage utility, WechatServiceButton, BulkEditAccountModal
- Tidy go.sum

P0:
- rpm_override 嵌入 Auth Cache Snapshot，消除每请求 DB 查询 (snapshot v6→v7)
- 429 RPM 响应返回 Retry-After 头（当前分钟剩余秒数）

P1:
- ClearAll 按钮直连 DELETE API，带 loading 防重复
- 新增 GET /admin/users/:id/rpm-status 管理员 RPM 用量查询端点

优化:
- checkRPM 从级联互斥改为并行取最严，user.rpm_limit 作为全局硬上限始终生效
- Override/Group 变更后自动失效 auth cache
- fail-open 语义不变，Redis 故障不阻塞业务

Why: channels with model pricing entries but no model mapping (e.g. azcc with
3 priced claude models, no mapping) were rendering as 未配置模型 in the
'Available Channels' page. The algorithm only iterated ModelMapping and
silently dropped any platform without a mapping entry.

Changes:
- channel.go: SupportedModels now unions mapping + pricing entries.
  For exact mapping src → target, pricing is looked up by target (the actually
  billed name), not by src.
- channel_available.go: ListAvailable enriches each entry with nil pricing
  via PricingService.GetModelPricing (global LiteLLM fallback) so the popover
  always shows a price.
- channel_service.go: NewChannelService takes *PricingService as 4th param.
- channel_test.go: rewrote 4 tests that froze the old mapping-only semantics;
  added pricing-only / mapping-target / target-missing coverage.

Add a DB-backed soft switch "available_channels_enabled" controlling
the user-facing /channels/available endpoint and sidebar entry. Default
to false (opt-in) — the feature stays invisible until an admin enables
it under Admin Settings > Features.

- domain_constants: SettingKeyAvailableChannelsEnabled
- settings_view: AllSettings/PublicSettings + AvailableChannelsEnabled
- setting_service: public+all read/write, seed default "false",
  GetAvailableChannelsRuntime helper (fail-closed on read error)
- admin setting_handler: UpdateSettingsRequest *bool + update branch
  + audit diff entry
- public setting_handler: expose via GET /api/v1/settings
- available_channel_handler: featureEnabled() guard — returns empty
  list after auth when disabled (401 precedes the feature check to
  preserve existing behavior)

Remove the admin-side "Available Channels" aggregate view — admins
already see full channel configuration (groups, pricing, model
mappings) in the channel edit dialog, making a read-only admin
aggregate view redundant. The user-side "可用渠道" remains.

Backend:
- Delete handler/admin/available_channel_handler.go (+ test)
- Drop AdminHandlers.AvailableChannel field and wire injection
- Remove /admin/channels/available route

Frontend:
- Delete views/admin/AvailableChannelsView.vue
- Drop /admin/available-channels router entry
- Strip AvailableChannel types + listAvailable from api/admin/channels.ts

Add a read-only aggregate view per channel: its linked groups and a
deterministic wildcard-free supported-model list with pricing details.

Backend
- service.Channel.SupportedModels(): combine ModelMapping keys with
  same-platform ModelPricing.Models; trailing "*" keys expand via
  pricing prefix match; platforms without a mapping produce no
  entries (intentional "no mapping = not shown" rule).
- Extract splitWildcardSuffix() shared with toModelEntry.
- Build a per-call pricing lookup map (platform+lowerName -> *pricing)
  to avoid O(N*M) scans in SupportedModels.
- ChannelService.ListAvailable() aggregates channels + active groups;
  filters out group IDs no longer active.
- Admin route GET /api/v1/admin/channels/available returns the full
  DTO (id, status, billing_model_source, restrict_models, groups,
  supported_models).
- User route GET /api/v1/channels/available applies three filters:
  Status==active, visible-group intersection, and platform filter
  on supported_models (prevents cross-platform leak when a channel
  links to both a user-accessible group and an inaccessible one on
  another platform). Response is a plain array (matches the
  /groups/available sibling shape). Field whitelist omits
  billing_model_source, restrict_models, ids, status, sort_order.

Frontend
- New /admin/available-channels and /available-channels views backed
  by a shared AvailableChannelsTable component (admin adds status +
  billing-source columns via slots).
- PricingRow extracted to its own SFC; SupportedModelChip references
  shared billing-mode constants in constants/channel.ts.
- Sidebar: new entry above "渠道管理" for admin; matching entry in
  user nav.
- i18n: zh + en coverage for both namespaces.

Tests
- SupportedModels: wildcard-only pricing skipped, prefix-matches-
  nothing, cross-platform bleed, case-insensitive dedup, empty
  platform mapping.
- ListAvailable: nil groupRepo, inactive-group-ID dropped, stable
  case-insensitive name sort.
- User handler: 401 on unauthenticated, visible-group intersection,
  platform filter on supported_models, JSON whitelist.
- Admin handler: full DTO including default BillingModelSource
  fallback.

Refs: issue #1729

Problem:
Upstream channels can reject monitor probes based on client fingerprint
(e.g. "only Claude Code clients allowed"). The monitor had no way to
customize the outgoing request to bypass such restrictions.

Solution:
Introduce reusable request templates that carry extra_headers plus an
optional body override; monitors reference a template and receive a
snapshot copy on apply. Template edits do NOT auto-propagate — users
must click "apply to associated monitors" to refresh snapshots, so a
bad template edit cannot instantly break all production monitors.

Data model (migration 112):
- channel_monitor_request_templates: id, name, provider, description,
  extra_headers jsonb, body_override_mode ('off'|'merge'|'replace'),
  body_override jsonb. Unique (provider, name).
- channel_monitors: +template_id (FK, ON DELETE SET NULL), +extra_headers,
  +body_override_mode, +body_override (the three runtime snapshot fields).

Checker (channel_monitor_checker.go):
- callProvider + runCheckForModel accept a CheckOptions carrying the
  snapshot fields. mergeHeaders applies user headers on top of adapter
  defaults (forbidden list: Host / Content-Length / Transfer-Encoding /
  Connection / Content-Encoding).
- buildRequestBody:
    off     -> adapter default body
    merge   -> shallow-merge over default; per-provider deny list
               (model/messages/contents) protects the challenge contract
    replace -> user body verbatim
- Replace mode skips challenge validation; instead HTTP 2xx + non-empty
  extracted response text = operational, empty = failed.
- 4 new unit tests cover all three modes + replace/empty-response case.

Admin API:
- /admin/channel-monitor-templates CRUD + /:id/apply (overwrite snapshot
  on all template_id=id monitors, returns affected count).
- channel_monitor request/response DTOs gain the 4 new fields.

Frontend:
- channelMonitorTemplate.ts API client.
- MonitorAdvancedRequestConfig.vue shared component for headers textarea
  + body mode radio + body JSON editor; used by both template and monitor
  forms.
- MonitorTemplateManagerDialog.vue: provider tabs, list/create/edit/
  delete/apply, live "associated monitors" count per row.
- MonitorFiltersBar: new 模板管理 button next to 新增监控.
- MonitorFormDialog: collapsible 高级 section with template dropdown
  (filtered by form.provider, clears on provider change) + embedded
  AdvancedRequestConfig. Picking a template copies its fields into the
  form (snapshot semantics mirrored on the client).
- i18n zh/en entries for all new copy.

chore: bump version to 0.1.114.32

明细只保留 1 天，超过 1 天聚合到新表 channel_monitor_daily_rollups（按
monitor_id/model/bucket_date 维度），聚合保留 30 天。两张表都用 SoftDeleteMixin
软删除（DELETE 自动改为 UPDATE deleted_at = NOW()）。

聚合 + 清理任务由 OpsCleanupService 的 cron 统一调度，与运维监控的清理共享
schedule（默认 0 2 * * *）和 leader lock。ChannelMonitorRunner 的 cleanupLoop
被移除，只保留 dueCheckLoop。

读取路径 ComputeAvailability* 改为 UNION 明细（今天 deleted_at IS NULL）+
聚合（过去 windowDays 天 deleted_at IS NULL），SUM(ok)/SUM(total) 自然加权
计算可用率，AVG latency 用 SUM(sum_latency_ms)/SUM(count_latency)。

watermark 表 channel_monitor_aggregation_watermark 单行（id=1），记录
last_aggregated_date，重启后从该日期 +1 继续聚合，首次为 nil 则从
today - 30d 开始回填，单次最多 35 天上限避免长事务。

raw SQL 的 ListLatestPerModel / ListLatestForMonitorIDs / ListRecentHistoryForMonitors
都补上 deleted_at IS NULL 过滤（SoftDeleteMixin interceptor 只对 ent query 生效）。

bump version to 0.1.114.28

GroupBadge 在 MonitorKeyPickerDialog 中复用平台主题色 + 倍率/专属倍率
（顺手优化）。

Settings:
- New "功能开关" tab between 通用设置 and 安全与认证
- ChannelMonitorEnabled toggle: runner skips scheduling when false,
  user-facing list returns empty
- ChannelMonitorDefaultIntervalSeconds (15-3600): pre-fills interval
  when creating a new monitor; each monitor can still override

Bug fix:
- ModelTagInput now commits pending input on blur, not just Enter/Tab.
  Previously clicking "save" with an un-Enter'd extra model would drop
  the value (DB stored extra_models=[] even when user typed entries).

Backend:
- domain_constants: SettingKeyChannelMonitor{Enabled,DefaultIntervalSeconds}
- SettingService.GetChannelMonitorRuntime: lightweight getter used by
  runner tick + user handler per-request (fail-open on DB error)
- Runner tickDueChecks: bails early when feature disabled
- ChannelMonitorUserHandler: checks feature flag before serving
- Comment on runner doc: scheduler state is implicit (every tick re-reads
  ListEnabled from DB), so CRUD ops on monitors self-maintain the schedule

Bump VERSION to 0.1.114.25

新增 admin「渠道监控」模块（参考 BingZi-233/check-cx），独立于现有 Channel 体系。
admin 配置 + 后台定时调用上游 LLM chat completions 健康检查 + 所有登录用户只读可见。

后端：
- ent: channel_monitor + channel_monitor_history（AES-256-GCM 加密 api_key）
- service 按职责拆分：service/aggregator/validate/checker/runner/ssrf
- provider strategy map 替代 switch（openai/anthropic/gemini）
- repository batch 聚合（ListLatestForMonitorIDs + ComputeAvailabilityForMonitors）消除 N+1
- runner: ticker(5s) + pond worker pool(5) + inFlight 防并发 + TrySubmit 防雪崩
  + 凌晨 3 点 cron 清理 30 天历史
- SSRF 防护：强制 https + 私网/loopback/云元数据 IP 拒绝（127/8、10/8、172.16/12、
  192.168/16、169.254/16、100.64/10、::1、fc00::/7、fe80::/10）+ DialContext
  在 socket 层防 DNS rebinding
- API key sanitize：擦除 url.Error 与上游响应 body 中的 sk-/sk-ant-/AIza/JWT 模式
- APIKeyDecryptFailed 标志位 + 单 monitor 路径检测，避免空 key 调用上游

handler:
- admin: CRUD + 手动触发 + 历史接口（api_key 脱敏）
- user: 只读列表 + 状态详情（去除 api_key/endpoint）
- ParseChannelMonitorID 共用 + dto.ChannelMonitorExtraModelStatus 共用

前端：
- 路由 /admin/channels/{pricing,monitor} + /monitor（用户只读）
- AppSidebar 父项 expandOnly 支持
- ChannelMonitorView 拆为 8 个子组件 + ChannelStatusView 拆出 detail dialog
- composables/useChannelMonitorFormat + constants/channelMonitor 共享
- i18n monitorCommon namespace 消除 admin/user 两 view 重复

合规：所有文件符合 CLAUDE.md（Go ≤ 500 行 / Vue ≤ 300 行 / 函数 ≤ 30 行）
CI: go build / gofmt / golangci-lint(0 issues) / make test-unit / pnpm build 全绿

wire_gen.go was calling NewSchedulerCache(redisClient) but wire.go had
been updated to register ProvideSchedulerCache(redisClient, config),
which reads SnapshotMGetChunkSize and SnapshotWriteChunkSize from config.
Without this fix, those config values were silently ignored.

- Restore gateway_cache.go to upstream (no lua embeds)
- Restore payment_order.go to upstream (use out_trade_no lookup)
- Restore payment_fulfillment.go to upstream (same reason)
- Add FeaturesConfig field and IsWebSearchEmulationEnabled to Channel
- Add applyAccountStatsCost wrapper function
- Add SettingKeyWebSearchEmulationConfig constant
- Add WebSearchEmulationEnabled to SystemSettings
- Add notify code rate limiting methods to EmailCache interface
- Remove AllowUserRefund references (ent schema not present)
- Fix duplicate import in payment_handler.go
- Fix wire_gen.go argument mismatches

WebSearch tri-state switch:
- Account-level web_search_emulation changed from bool to tri-state
  string: "default" (follow channel) / "enabled" / "disabled"
- shouldEmulateWebSearch checks channel config when account is "default"
- SQL migration converts old bool values
- Frontend select replaces toggle in Edit/CreateAccountModal

Account stats pricing:
- resolveAccountStatsCost uses upstream model (post-mapping) for matching
- Priority: custom rules → model pricing file (when toggle on) → default
- Custom rules always configurable, independent of toggle
- Account ID field changed to searchable selector filtered by platform
- Description updated to reflect new behavior

Quota notification cache fix:
- CheckAccountQuotaAfterIncrement fetches real-time account from DB
- Reconstructs pre-increment usage for accurate threshold crossing detection
- New AccountQuotaReader interface (minimal: GetByID only)

Usage tooltip:
- Per-request/image billing shows per-request price instead of $0 token price
- Token billing continues to show input/output price per million tokens

- Fix GetByKeyForAuth not selecting balance notify fields (notifications
  never triggered in gateway path)
- Fix provider-level ProxyURL never resolved: inject ProxyRepository into
  SettingService, resolve proxy URLs when building Manager
- Fix admin manual balance adjustment not updating total_recharged
- Add threshold_type input validation (reject invalid values)
- Fix user threshold_type inheritance: custom threshold defaults to "fixed"
  instead of inheriting global type (prevents $5 being treated as 5%)
- Add try-catch for clipboard.writeText (fails on non-HTTPS)
- Add SetTotalRecharged to user Update for admin balance operations

- User balance low notification: email alert when balance drops below
  configurable threshold (user email + verified extra emails)
- Account quota notification: broadcast email to admin-configured
  recipients when daily/weekly/total quota usage exceeds alert threshold
- Admin settings: global enable/disable, default threshold, quota
  notification email list (Email Settings tab)
- User profile: enable/disable, custom threshold, add/remove extra
  notification emails with verification code flow
- Account quota: per-dimension alert toggle and threshold in quota
  control card
- Trigger logic: first-crossing only (old >= threshold && new < threshold
  for balance; old < threshold && new >= threshold for quota), naturally
  prevents duplicate notifications without Redis dedup

Backend:
- Extract YuanToFen/FenToYuan to payment/amount.go using shopspring/decimal
- Require alipay publicKey in config validation
- Fix wxpay webhook response to return JSON per V3 spec
- Remove wxpay certSerial fallback to publicKeyId
- Define magic strings as named constants in wxpay/alipay providers
- Add slog warning for wxpay H5→Native payment downgrade
- Make EncryptionKey validation return error on invalid (non-empty) key
- Make decryptConfig propagate errors instead of returning nil
- Add idempotency check in doBalance to prevent stuck FAILED retries

Frontend:
- Fix dashboard currency symbol from $ to ¥
- Fix AdminPaymentPlansView any type to proper SubscriptionPlan type
- Make quick amount buttons follow selected payment method limits
- Center help image with larger height and text below

Add a full payment and subscription system supporting EasyPay (Alipay/WeChat),
Stripe, and direct Alipay/WeChat Pay providers with multi-instance load balancing.

Restore account_usage_service.go, antigravity_gateway_service.go,
antigravity_credits_overages.go and its test to upstream/main state.
These credits balance precheck changes were accidentally included
during cherry-pick of channel management commits.

- Parse candidatesTokensDetails from Gemini API to separate image/text output tokens
- Add image_output_tokens and image_output_cost to usage_log (migration 089)
- Support per-image-token pricing via output_cost_per_image_token from model pricing data
- Channel pricing ImageOutputPrice override works in token billing mode
- Auto-fill image_output_price in channel pricing form from model defaults
- Add "channel_mapped" billing model source as new default (migration 088)
- Bills by model name after channel mapping, before account mapping
- Fix channel cache error TTL sign error (115s → 5s)
- Fix Update channel only invalidating new groups, not removed groups
- Fix frontend model_mapping clearing sending undefined instead of {}
- Credits balance precheck via shared AccountUsageService cache before injection
- Skip credits injection for accounts with insufficient balance
- Don't mark credits exhausted for "exhausted your capacity on this model" 429s

- 定价查找支持通配符(suffix *)，最长前缀优先匹配
- 模型限制(restrict_models)同样支持通配符匹配
- OpenAI 网关接入渠道映射/BillingModelSource/模型限制
- 按次/图片计费模式创建时强制要求价格或层级(前后端)
- 用户使用记录列表增加计费模式 badge 列

- GatewayService/OpenAIGatewayService 注入 ModelPricingResolver
- RecordUsage 从旧路径迁移到 CalculateCostUnified（支持 per_request/image 模式）
- 无渠道时自动回退旧路径，保持原有行为
- 长上下文双倍计费仅在无渠道定价时生效
- CostBreakdown 新增 BillingMode 字段，使用日志记录实际计费模式
- 模型限制错误改为与"无可用账号"相同的 503 响应

- 新增 GET /admin/channels/model-pricing?model=xxx API
- 从 BillingService 查询 LiteLLM/Fallback 默认定价
- 前端添加模型时自动查询并填充价格（$/MTok）
- 仅在所有价格字段为空时才自动填充，不覆盖手动配置

Cherry-picked from release/custom-0.1.106: a9117600

When an antigravity account returns 500 "Internal error encountered."
on all 3 retry attempts, increment a Redis counter and apply escalating
penalties:
- 1st round: temp unschedulable 10 minutes
- 2nd round: temp unschedulable 10 hours
- 3rd round: permanently mark as error

Counter resets on any successful response (< 400).

新增功能：
- 新增 TLS 指纹 Profile CRUD 管理（Ent schema + 迁移 + Admin API + 前端管理界面）
- 支持账号绑定数据库中的自定义 TLS Profile，或随机选择（profile_id=-1）
- HTTPUpstream.DoWithTLS 接口从 bool 改为 *tlsfingerprint.Profile，支持按账号指定 Profile
- AccountUsageService 注入 TLSFingerprintProfileService，统一 usage 场景与网关的 Profile 解析逻辑

代码优化：
- 删除已被 TLSFingerprintProfileService 完全取代的 registry.go 死代码（418 行）
- 提取 3 个 dialer 的重复 TLS 握手逻辑为 performTLSHandshake() 共用函数
- 修复 GetTLSFingerprintProfileID 缺少 json.Number 处理的 bug
- gateway_service.Forward 中 ResolveTLSProfile 从重试循环内重复调用改为预解析局部变量
- 删除冗余的 buildClientHelloSpec() 单行 wrapper 和 int64(e.ID) 无效转换
- tls_fingerprint_profile_cache.go 日志从 log.Printf 改为 slog 结构化日志
- dialer_capture_test.go 添加 //go:build integration 标签，防止 CI 失败
- 去重 TestProfileExpectation 类型至共享 test_types_test.go
- 修复 9 个测试文件缺少 tlsfingerprint import 的编译错误
- 修复 error_policy_integration_test.go 中 handleError 回调签名被错误替换的问题

1. accounts/check 补全 plan_type：当 id_token 缺少 plan_type（如 Mobile RT），
   自动调用 accounts/check 端点获取订阅类型
2. orgID 精确匹配账号：从 JWT 提取 poid 匹配正确账号，避免 Go map
   遍历顺序随机导致 plan_type 不稳定
3. RT 刷新时设置隐私：调用 disableOpenAITraining 关闭训练数据共享，
   结果存入 extra.privacy_mode，后续跳过重复设置
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

## Problem

When a proxy is unreachable, token refresh retries up to 4 times with
30s timeout each, causing requests to hang for ~2 minutes before
failing with a generic 502 error. The failed account is not marked,
so subsequent requests keep hitting it.

## Changes

### Proxy connection fast-fail
- Set TCP dial timeout to 5s and TLS handshake timeout to 5s on
  antigravity client, so proxy connectivity issues fail within 5s
  instead of 30s
- Reduce overall HTTP client timeout from 30s to 10s
- Export `IsConnectionError` for service-layer use
- Detect proxy connection errors in `RefreshToken` and return
  immediately with "proxy unavailable" error (no retries)

### Token refresh temp-unschedulable
- Add 8s context timeout for token refresh on request path
- Mark account as temp-unschedulable for 10min when refresh fails
  (both background `TokenRefreshService` and request-path
  `GetAccessToken`)
- Sync temp-unschedulable state to Redis cache for immediate
  scheduler effect
- Inject `TempUnschedCache` into `AntigravityTokenProvider`

### Account failover
- Return `UpstreamFailoverError` on `GetAccessToken` failure in
  `Forward`/`ForwardGemini` to trigger handler-level account switch
  instead of returning 502 directly

### Proxy probe alignment
- Apply same 5s dial/TLS timeout to shared `httpclient` pool
- Reduce proxy probe timeout from 30s to 10s

复用 GroupCapacityService，在 admin 分组列表中添加容量列，
显示每个分组的实时并发/会话/RPM 使用量和上限。
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>