E2: Deeper healthcheck (report degraded on stalled collection)

[LarsLaskowski]

Epic: Cross-cutting / smaller items

Suggested extra label: ops.

Why

GET /healthz returns a static ok (server.go:76 → handleHealthz); it stays green even if the collector goroutine has stalled and the metrics are stale. systemd / uptime monitors cannot detect that failure mode.

Scope

Optionally report unhealthy (non-200) when the latest snapshot is older than a configurable staleness bound. The snapshot already carries a Timestamp (collector.go:138 Snapshot()), so freshness is now - Snapshot().Timestamp compared against a multiple of the fast poll interval.

Reuse & touch points

• internal/httpapi/handlers.go — handleHealthz.
• MetricsProvider already exposes Snapshot() (server.go:19).
• Optional config healthz_max_staleness.

API & config impact

/healthz may now return non-200 (document this); keep it auth-exempt as today (handlers_test.go:160).

Acceptance criteria

• A fake provider with a stale timestamp yields a non-200 response.
• A fresh snapshot yields 200 ok.