fix: MQTT connection flapping + comprehensive bug fixes

[eman]

Summary

Comprehensive bug fix PR addressing the MQTT connection flapping issue plus 9 additional bugs found during a full code audit.

MQTT Connection Flapping (Original Fix)

Problem

MQTT connections flap endlessly: connect → disconnect after 1-7s → reconnect → repeat. All entities become unavailable in Home Assistant.

Root Cause

When _active_reconnect() creates a new MqttConnection, the old connection is never closed. The old AWS IoT SDK connection's built-in auto-reconnect eventually succeeds, creating two active connections with the same client ID. AWS IoT kicks one off → on_connection_interrupted fires → another reconnection → infinite loop.

Fix

• Add MqttConnection.close() method that unconditionally tears down the SDK connection regardless of _connected state
• _active_reconnect(): close old connection before creating replacement
• _deep_reconnect(): use close() unconditionally instead of checking is_connected

Additional Bug Fixes (Code Audit)

Thread Safety

• ensure_device_info_cached: Use loop.call_soon_threadsafe() for Future resolution from AWS SDK callback thread (was calling future.set_result() directly from non-asyncio thread — race condition / potential crash)

MQTT Reconnection

• Config validation: Clamp deep_reconnect_threshold to minimum of 1 to prevent ZeroDivisionError in reconnection backoff modulo operation

Diagnostics

• Counter bug: Increment total_reconnect_attempts on each connection drop (was always 0)
• JSON serialization: Replace float('inf') default for shortest_session_seconds with None in to_dict() (was breaking JSON serialization)

Events

• Future binding: Use asyncio.get_running_loop().create_future() instead of bare asyncio.Future() in wait_for()

Encoding

• Unit-aware temp validation: build_reservation_entry() now uses correct defaults per unit system (35-65°C metric, 95-150°F US). Was hardcoded to Fahrenheit values, causing valid Celsius inputs to fail validation.
• Malformed data handling: decode_reservation_hex() logs a warning on trailing bytes instead of silently discarding them

Periodic Requests

• Unknown request type: Log error and stop task on unknown PeriodicRequestType instead of silently doing nothing

Cache

• Memory leak: Purge expired entries from device_info_cache during get_all_cached() instead of only filtering results

Testing

• All 486 tests pass (463 existing + 23 new)
• New test files: test_mqtt_reconnection.py, test_bug_fixes.py
• Verified MQTT stability in live Home Assistant docker container

[Copilot]

Pull request overview

This PR addresses MQTT connection flapping in the MQTT client by ensuring old AWS IoT SDK connections are torn down before creating replacement connections, and bundles several additional audit-driven bug fixes across diagnostics, events, encoding, periodic requests, and caching.

Changes:

• Add an unconditional MqttConnection.close() teardown and use it during active/deep reconnect flows to prevent competing connections with the same client ID.
• Fix several correctness issues found during audit (thread-safe future resolution from SDK thread, JSON-safe diagnostics serialization, reconnection config validation, event-loop-bound futures).
• Add new tests covering reconnection cleanup and the additional bug fixes.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/test_mqtt_reconnection.py	New tests for MqttConnection.close() and reconnection cleanup behavior.
tests/test_bug_fixes.py	New tests covering config clamping, diagnostics metrics serialization, event wait_for(), reservation encoding/decoding, and cache purge behavior.
src/nwp500/mqtt/utils.py	Clamp deep_reconnect_threshold to a minimum of 1 during config initialization.
src/nwp500/mqtt/periodic.py	Log and stop the periodic loop on unknown PeriodicRequestType.
src/nwp500/mqtt/diagnostics.py	Make metrics JSON-friendly (inf → None) and increment reconnect attempts on drops.
src/nwp500/mqtt/connection.py	Introduce close() that disconnects regardless of _connected state and clears internal state to prevent SDK auto-reconnect interference.
src/nwp500/mqtt/client.py	Close old connection manager before rebuilding during reconnection; make device-info caching future resolution thread-safe.
src/nwp500/events.py	Bind wait_for() futures to the running event loop via get_running_loop().create_future().
src/nwp500/encoding.py	Add warning on trailing bytes in reservation hex decode; make reservation temp validation unit-system aware.
src/nwp500/device_info_cache.py	Purge expired entries during get_all_cached() to prevent cache growth.