feat(spine): red probes and error spikes now REACH the owner #5471

The spine's transition detector existed and worked — and delivered every alert to MONITOR_ALERT_WEBHOOK_URL, an env var nothing sets in production. Detection without delivery: the spine could see, but had no mouth.

New spine-alert-fanout:

• green→red edges, batched — a full outage flips ~14 checks in one tick and produces ONE message, not fourteen
• recovery notices when checks return to green (an outage email with no recovery email leaves the reader assuming the worst)
• new-error spikes — 3+ novel platform_errors fingerprints inside 10 minutes, the signature of a broken deploy even while every endpoint still answers 200; 1-hour cooldown

Every site admin gets an in-app notification AND an email (Resend when configured; degrades to log). Nothing in the path can throw into the monitor tick. Unit tests exercise batching, edge-triggering, recovery, and the fresh-DB case via the injected fanOut dep.

Part 2 of the reliability build (1 = external heartbeat on the GitHub mirror, live; next: deploy rollback, CI test gate, synthetic journeys).

🤖 Generated with Claude Code