fix(job-retry): fail open instead of dropping the job when isJobQueued errors

[atsikham]

Fixes #5423

Summary

checkAndRetryJob had no error handling around its call to isJobQueued. Any error from it (a transient GitHub API error, a rate limit, an unsupported event type) propagated out of the function, and the lambda handler only logs and swallows that error — the SQS message is marked processed and never redelivered, so the job's retry is dropped for good, and the RetryJob metric for that attempt is never recorded either.

Changes

• isJobQueued now runs inside a try/catch, mirroring the equivalent check already in scale-up.ts:
  • On UnsupportedEventError, the retry is skipped (that error can never resolve itself, retrying won't help).
  • On any other error, the job is assumed to still be queued and the retry is published anyway — a transient error is not evidence the job stopped needing a runner.

Note on overlap with #5422

This touches the same lines as #5422 (which makes the retry lambda respect ENABLE_JOB_QUEUED_CHECK) — they're different bugs (a missing config check vs. missing error handling) found independently, so kept as separate PRs rather than bundled. Whichever merges first, the other will likely need a small rebase.

Test plan

• yarn test — two new cases in job-retry.test.ts: a transient isJobQueued error still publishes the retry and records the metric (fail-open), and an unsupported event type skips the retry without publishing. Full control-plane suite passes (358 tests).
• yarn lint / yarn format
• tsc --noEmit