v2.0.3

Fixed

Commands no longer silently no-op on stale cached state (#27).

Since v1.12.14 a 504 from the Gardena/Husqvarna gateway triggered a retry-until-confirmed loop that polled the cached coordinator state for confirmation. The very first check ran against whatever was already in memory — which, after a top-of-the-hour 504, could still be a "mowing-ish" activity left over from the previous cycle (OK_SEARCHING and OK_LEAVING are both in the mowing set, and the cache can be stale). The command was then reported as a success at INFO level, no retry happened, the mower never actually started, and — because HA's default log level is WARNING — nothing was logged.

Confirmation now requires a fresh WebSocket push for that specific device that arrives after the command was sent (a per-device monotonic push marker captured before each attempt), in addition to the expected-state check. Stale cache can no longer be mistaken for a landed command: an unconfirmed command now correctly retries and, if still unconfirmed, surfaces a visible command_timeout instead of silently doing nothing. Applies to both the Gardena and Automower command paths.

Full test suite (636) passing; ruff and mypy --strict clean.