Health Check

Aggregated health for the gateway and all workers it routes to.

• Free: only does internal HTTP GETs to each worker's /health (no model calls, no external traffic). Internal traffic on Fly.io flycast and Railway private networking is free.
• Cached for ~10s (HEALTH_CACHE_TTL) so frequent polling stays cheap.
• Pass ?refresh=1 to force a fresh probe. Returns 200 always (gateway itself is up); inspect the JSON body for per-worker status. Top-level status is "ok" / "degraded" / "down".

Public response is sanitised — it never exposes internal .railway.internal hostnames or other infra fingerprints. Callers presenting the gateway's X-Internal-Secret header receive the unredacted snapshot (URLs, model aliases, machine state).