GitLab

* Always try and start gpg-agent, with --use-standard-socket it will try and use a standard socket directory. It won't start multiple agents if agent is already running. In addition, XDG_RUNTIME_DIR isn't always set

* ssh socket if broken if --daemon is run again, so onky start if we don't have a socket

* Removed unnecessary allocation of GPG_SSH_AUTH_SOCK

$GNUPGHOME variable was used incorrectly and caused a grep error when
set.

The gpg-agent plugin did not work for gpg versions above or equal to
2.1 because of the `--write-env-file` option deprecation. This new version
works fine and also enables the ssh-agent support only if it is enabled in the
gpg-agent config file.

If using the gpg-agent with --enable-ssh-support, the SSH_AUTH_SOCK and
SSH_AGENT_PID environment variables need to be exported once sourced
from GPG_ENV.  Otherwise, we get no benefit from the persisting these
values to GPG_ENV; subsequent openned terminals don't see the existent
gpg-agent as a process for an SSH daemon.

The GPG_ENV file is sourced before doing the gpg-connect-agent check,
but this file (unlike the SSH_ENV file) doesn't export GPG_AGENT_INFO,
so the check always fails.  This results in new gpg-agents continuously
being spawned.

All this commit does is put in the single export to fix the problem.

Don't just overwrite the environment. First check for a running agent (an
x-session might have one running). If no agent is found, source the
environment and check again using those settings. If again no agent is
found, start a new instance.

Based on ssh-agent plugin.