Give every automation a dead-man switch

Automations fail quietly. The cron job that stopped running last Tuesday does not send an error, it sends nothing.

The fix is old and cheap: every job pings a heartbeat endpoint when it finishes. A separate tiny job checks for missing heartbeats and messages you. Now the alert fires on absence, not on error.

• One heartbeat per job, named after the job.
• Grace period equal to two run intervals.
• The checker itself is watched by a second checker on a different machine. Yes, really.