Agent registration drifts from container reality (stale IP, stale running status)

[jaylfc]

Found while live-verifying the #97 resume fix on the dev Pi.

Agent naira (hermes, LXC):

• Registered host is 10.26.37.174, but when its container starts it now comes up on 192.168.7.5 (bridge/subnet changed since registration). Every controller call (prepare-for-shutdown, resume, chat) targets the dead address.
• The agent showed status: running in /api/agents while its container was actually STOPPED, so the UI presents a live agent that is not there.

Consequences: the boot-time resume retries the wrong address and correctly ends with the still-paused warning, but the underlying cause is invisible; users see a paused, apparently-running agent that never comes back.

Suggested direction:

1. Reconcile status with the container runtime state (incus/docker query) rather than trusting the stored value.
2. Refresh the registered host from the container's current IP on start (or resolve by container name at call time instead of storing a raw IP).
3. Surface address drift explicitly in the Agents app when the registered IP does not match the container's.

Not urgent for tonight; the #97 warning makes the symptom visible instead of silent.

[jaylfc]

Landed via #1884. Worker host_lan_ip/url/hardware now refresh on every heartbeat, so the cluster follows DHCP moves and LXC restarts instead of routing to stale addresses. Thanks @hognek.