Improve reliability of scheduled event publish cron (~every 5 minutes)

[akhilk999]

Summary

GitHub Actions runs .github/workflows/publish-scheduled-events.yml on */5 * * * * to call /api/cron/publish-events and publish due scheduled events. The job is intended to poll about every 5 minutes, but GitHub's scheduler is best-effort, not guaranteed.

Observed behavior

• Manual workflow_dispatch runs work after the middleware /api/cron allowlist fix.
• Automatic schedule runs do fire, but not reliably every 5 minutes.
• Example: after a successful schedule run at 2026-07-17T00:06:55Z, roughly six expected */5 ticks (~00:10–00:35 UTC) did not produce additional schedule runs (~30 minutes with no automatic poll).

Impact

• Officers can schedule an event for time T, but it may publish several minutes (or longer) after publish_at.
• Empty polls ({"published":0}) are fine when they do run; the problem is missed / delayed schedule ticks.

Context / constraints

• Vercel Hobby only allows ~1 cron/day, so we moved polling to GitHub Actions.
• Keepalive job was added to reduce 60-day public-repo schedule dormancy risk; that does not fix tick reliability.

Possible directions

1. Monitor schedule gaps (alert if no schedule run for N minutes).
2. Use an external scheduler (e.g. cron-job.org) hitting the same authenticated endpoint.
3. Move to Vercel Pro cron if we want platform-side minute accuracy.
4. Document expected "publish within ~5–15+ minutes after publish_at" for officers.

Acceptance criteria

• Documented expected publish delay / reliability for scheduled events
• Chosen mitigation (monitor, external cron, and/or Vercel Pro) implemented or explicitly deferred
• Verification that automatic publishes happen close enough to publish_at for CSA ops needs