Fix HapMonitor sockets never reconnecting after same-port HAP instance restart

[Copilot]

When a paired HAP instance restarts on the same port with identical mDNS TXT records (md, port, c# unchanged), the monitor socket dies permanently. The bonjour up handler gates refreshMonitorConnection on a port/name/config diff — none fire on an identical restart — and HapMonitor has no self-healing logic on socket close.

Changes

src/monitor.ts — self-healing reconnect loop

• On socket.close, if !_stopped, schedules _scheduleReconnect(instance) with exponential backoff: 2 s initial, doubles per failure, capped at 30 s, +0–1 s jitter
• finish() sets _stopped = true and clears all pending timers before destroying sockets — intentional teardown does not trigger reconnects
• refreshMonitorConnection() cancels any pending reconnect timer before creating a new connection (prevents duplicate reconnects when mDNS and the backoff timer race)
• Null-guards the socket in refreshMonitorConnection for the case where the initial connectInstance failed before assignment
• Adds isInstanceConnected(username): boolean — queries live socket state without exposing internals

src/index.ts — fast reconnect path on mDNS re-announcement

• In the bonjour up handler, adds an else if branch: when an existing instance is re-announced unchanged but isInstanceConnected() is false, calls refreshMonitorConnection immediately rather than waiting for the backoff timer

} else if (this.hapMonitor && !this.hapMonitor.isInstanceConnected(instance.username)) {
  // Same port/name/config but socket is dead (e.g. same-port restart) — reconnect immediately
  this.hapMonitor.refreshMonitorConnection(this.instances[existingInstanceIndex])
}

The two fixes are complementary: the backoff loop handles the case where mDNS re-announcement is missed or delayed; the index.ts fast path fires an immediate reconnect as soon as mDNS confirms the bridge is back.