fix: reap orphan dashboard daemon on version-skew restart
The dashboard daemon could keep serving a stale version after a plugin upgrade — the self-heal restarted the service job but never reaped an orphaned dashboard-server.py holding the port (a prior label, a manual launch, or another runtime's daemon on the same port). Observed live: a daemon pinned to an old version for 16h while the new version was installed.
Fixes
- Reap the orphaned port-holder before restart (own daemon only, never a foreign process).
api/healthnow reportsversion; the restart is verified as landed via a bounded, identity-checked probe (None= safe degrade, never a false stale).- Honest escalation + logging — no more false "Auto-updated" when the process never changed.
Hardening (from adversarial review)
- Do-no-harm gate so escalation never reaps a daemon it can't replace (Codex /
daemon_disabledusers no longer left daemon-less). - Idempotency guard defuses a concurrent-SessionStart restart flap.
Full suite: 306 passed.