feat: on-demand refresh — reload service + auto-refresh on container events

[lebdim]

Summary

Two related improvements that make monitor_docker data update on demand instead of only every scan_interval:

1. New monitor_docker.reload service — force an immediate refresh of all monitored Docker data (container state, stats and engine info).
2. Auto-refresh on container events — the Docker event stream now triggers an immediate refresh on start/stop/die/kill/pause/unpause/restart.

Motivation

The sensors are push-based (should_poll=False) and the background poll loops await asyncio.sleep(scan_interval) between fetches, so there was previously no way to refresh on demand short of restarting Home Assistant (homeassistant.update_entity is a no-op on these entities). Also, _run_docker_events only acted on create/destroy/rename and merely logged start/stop, so a plain start/stop was reflected only on the next poll.

This is useful for:

• Dashboards/automations that need up-to-date container state right after an action.
• Setups that raise scan_interval (to reduce recorder churn) but still want a manual "Refresh now" button and instant reaction to start/stop.

Note: the reload service registration was already stubbed out in the code (commented async_setup_reload_service + an empty async_reset_platform). This implements it via a lightweight refresh rather than a full platform reload.

Implementation

• A shared asyncio.Event per DockerAPI, passed to each DockerContainerAPI. The plain interval asyncio.sleep(self._interval) in both the docker-info loop and each container loop is replaced by _sleep_or_refresh() = asyncio.wait_for(event.wait(), timeout=interval).
• request_refresh() does event.set() + event.clear(), waking every current waiter exactly once so they fetch immediately and push fresh values to the existing entities (no teardown/re-add; entity IDs and history preserved).
• The monitor_docker.reload service calls request_refresh() on every configured instance.
• _run_docker_events now calls request_refresh() on container start/stop/die/kill/pause/unpause/restart actions.
• Retry sleeps are untouched; the container loop falls back to a plain sleep when no event is provided (backwards-compatible).

Testing

Live Docker host (25 containers, scan_interval: 60, allinone):

• monitor_docker.reload appears in Developer Tools → Actions; calling it refreshes container sensors within ~1s (verified a container's CPU% changing off-cycle) vs. no change without it.
• Starting/stopping a container triggers an immediate refresh via the event stream.
• No teardown; entities and history intact; no errors in the log.

[lebdim]

cc @ualex73 — this implements the monitor_docker.reload service that was previously stubbed out (the commented async_setup_reload_service + empty async_reset_platform). It uses a lightweight on-demand refresh (shared asyncio.Event waking the poll loops) rather than a full platform reload, so entities/history are preserved. Happy to adjust naming or approach if you'd prefer. Thanks for the integration!