v0.16.3 - FCM listener watchdog

Stable promotion of v0.16.3-beta.1

Fixed

• FCM listener no longer reconnects after 3 sequential transport errors (#3) — the upstream firebase_messaging client used to abort its receiver and never reconnect, leaving the doorbell camera black: auto_on answered call_starting but the push carrying the signaling payload never arrived.
  • FcmPushClient now uses FcmPushClientConfig(abort_on_sequential_error_count=None) so the library keeps retrying with its own backoff.
  • A 60s watchdog (async_track_time_interval) revives the listener if is_started flips to False. ensure_running() is serialised by an asyncio.Lock so overlapping ticks cannot spawn parallel FcmPushClient instances while a slow start() is still handshaking.
  • The notification grace period is intentionally not re-armed on revival — the existing _processed_notifications dedup deque already filters re-deliveries, and re-arming the blackout could drop a real doorbell ring landing during the window.

Soak summary (24 h on production HA running v0.16.3-beta.1)

• 0 Shutting down push receiver due to N sequential errors log lines.
• 3 clean self-reconnects from the underlying firebase_messaging client.
• 0 watchdog revivals (listener never died).
• Camera preview verified end-to-end.

PRs

• #4 — fix
• #5 — release