v0.4.3 - survive WiFi drop-offs (zombie-client fix)

Critical robustness fix. If the controller dropped off WiFi while connected, the resulting OSError (errno 113, Host unreachable) killed the read loop WITHOUT triggering a reconnect - the integration became a zombie (stale entities, keep-alive silently failing, morning automation erroring) until HA was restarted.

Now ANY socket-level error (reset, timeout, host-unreachable) is treated as connection loss and hands over to the normal reconnect loop, so the integration recovers automatically as soon as the controller rejoins the network.

Highly recommended if your controller has flaky WiFi. Everything from v0.4.2 included. Regression-tested (54 tests).