Add a health endpoint and basic metrics to the kernel daemon

[sirtimid]

There is no operational visibility. grep -rn "prometheus|healthz|opentelemetry|metrics" packages/*/src returns nothing.

For anyone running this as a service, that means no way to answer "is it up", "is it keeping up", or "is it stuck" other than reading daemon.log. Combined with the run loop failing silently, a wedged kernel is currently undetectable without inspecting the log by hand.

Proposed:

• A health endpoint reflecting actual liveness — specifically including run-loop state, not just process liveness.
• Basic counters and gauges: deliveries processed, run-queue depth, crank duration, active vats, remote connections, retransmissions, revocations.
• A format something standard can scrape. OpenTelemetry or a Prometheus text endpoint; either is fine, but it should not be bespoke.

Depends on the run-loop health issue — a health endpoint that reports healthy while the run loop is dead is worse than none.