Fix HTTP state reconciliation when MQTT is active

[dhia-e-rzig]

Summary

• keep periodic Appliance.System.All polling active when HTTP is the selected protocol, even if an auxiliary MQTT connection is receiving pushes
• preserve polling suppression when MQTT is both selected and active
• add an MSG200 regression test for reconciling a missed per-channel garage-door push

Root cause

NamespaceHandler.async_poll_all() used _mqtt_active alone to decide that all state updates would arrive by push. With HTTP selected and cloud MQTT connected, a dropped or incomplete multi-channel push therefore disabled the HTTP reconciliation path too. One MSG200 channel could remain stale indefinitely even though a direct Appliance.System.All response contained the correct state.

The guard now checks both curr_protocol and _mqtt_active, so HTTP remains the authoritative polling transport while MQTT can still provide low-latency auxiliary pushes.

Related context: #538

Tests

• pytest tests/test_config_entry.py::test_device_http_poll_with_mqtt_active -q
• pytest tests/test_config_entry.py -q
• full CI suite: 36 passed
• black --check on changed files

[krahabb]

Hello @dhia-e-rzig,
Thank you for submitting this. Allow me to regain focus on meross_lan development since I was a bit off for some time ;)
I'll check this asap.