Follow-up to t-3390. lib/reboot-queue.ts exposes reclaimStuckExecuting(maxMs) which marks orphan executing requests as failed.

Add a cron job (every 30 min or hourly) that POSTs to a small wrapper endpoint (or use an isolated agent turn) calling this. Without it, a crashed reboot mid-execution leaves the queue stuck.

Spec:

• Cron: every 30 min
• Action: call reclaimStuckExecuting(5 * 60 * 1000) via either a new /api/reboot-queue/janitor POST endpoint OR an isolated agentTurn that runs a small node one-liner.
• Alert via WhatsApp relay if reclaimed > 0 (something went wrong).

No reboot needed (read-only / data-only change for cron config).