OSS + Enterprise Sentry support

[acunniffe]

This PR introduces support for logging performance metrics + errors to Sentry compatible endpoints

OSS releases will have telemetry enabled by default starting in 1.0.15. There's a flag in config.json to disable it

   "telemetry_oss": "off"

Enterprises can set their own DSN using

   "telemetry_enterprise_dsn": "<dsn>"

Or by building from source and setting SENTRY_ENTERPRISE=<dsn> cargo build

--

what's logged?

• Any errors from a pre commit or post commit hook (should not include user data).
• Performance violations. if we're outside our target overhead for wrapping git commands, the timings for pre/post commit hooks will be logged.
• operating system, version
• remote URLs of repos

--

performance impact

because git-ai is a git wrapper and does not want to introduce any major overhead to a git command. So we're trading realtime logs for performance. Logs are buffered in an append-onlyai/log file (one per pid to prevent locking). Logs are then sent to the server only when you do a git fetch or git push. A background process is enqueued to mop up the logs and send them off without slowing those commands down.

[acunniffe]

@svarlamov let's sync on defining our internal targets.

[svarlamov]

Sounds good. The biggest thing I noticed in playing around with it is that we probably need to set a floor on the 10% (at least for now). So like max(10% or 70ms)

➜  git-ai git:(feat/sentry) GIT_AI_DEBUG_PERFORMANCE=1 git commit
On branch feat/sentry
Your branch is up to date with 'origin/feat/sentry'.

nothing to commit, working tree clean
[git-ai (perf)] ᕽ Performance target violated for command: commit. Total duration: 51ms, Git duration: 7ms. Pre-command: 44ms, Post-command: 0ms.

[git-ai-cloud-dev]

you  ████████████████████████░░░░░░░░░░░░░░░░ ai
      61%                                                                                                         39%

AI code tracked with git-ai

[svarlamov]

Everything looks good. Left some thoughts on the thresholds