fix(gateway): refresh MQTT token on broker auth-failure to self-heal expired JWT

[mgazza]

Problem

The gateway MQTT access token is a short-lived (24h) JWT. When it expires, GatewayMQTT._mqtt_loop reconnects forever with the same rejected token — the broker returns CONNACK reason code 134 ("Bad user name or password") on every attempt — so PredBat silently loses all gateway control until the AppDaemon pod (and its rendered config) is rebuilt.

The existing proactive refresh (_check_token_refresh) runs from the housekeeping run() loop, which does not rescue a listener already stuck in the reconnect loop. Observed in production across multiple gateway instances: every instance lost control ~24h after its last token render, with predbat.status reporting Inverter 0 write to charge_rate/scheduled_discharge_enable failed (a side effect — batpred could not publish anything).

Fix

Make the reconnect loop self-healing: when the broker rejects authentication, force a token refresh before reconnecting, instead of retrying the dead token.

• _is_auth_failure(error) — classify CONNACK 134/135 auth rejections vs ordinary drops (e.g. "Disconnected during message iteration", network errors), so we only refresh when a new token can actually help.
• _maybe_refresh_on_auth_error(error) — force a refresh only for auth failures.
• _do_token_refresh() / _apply_refresh_response() — extracted from _check_token_refresh so the proactive near-expiry path and the reconnect path share one refresh implementation.
• Wired into _mqtt_loop's exception handler (before the reconnect backoff sleep).

No behaviour change on the happy path or for non-auth disconnects.

Testing

New apps/predbat/tests/test_gateway_token_refresh.py (12 tests): auth-failure classification, response application (epoch + ISO expiry, failure leaves token unchanged), and that a refresh is forced only on auth failure. Existing gateway tests remain green.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates the GatewayMQTT component to self-heal when the MQTT broker rejects authentication (e.g., expired 24h JWT) by forcing a token refresh inside the reconnect loop, preventing infinite reconnect attempts with a dead token.

Changes:

• Refresh MQTT token on broker auth-failure during _mqtt_loop reconnect handling.
• Refactor token refresh logic into shared helpers (_is_auth_failure, _maybe_refresh_on_auth_error, _apply_refresh_response, _do_token_refresh) used by both proactive refresh and reconnect paths.
• Add a new test module for auth-failure classification and refresh response handling, plus cspell dictionary entries.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
apps/predbat/gateway.py	Adds auth-failure detection and forces MQTT JWT refresh before reconnecting; refactors refresh implementation into helper methods.
apps/predbat/tests/test_gateway_token_refresh.py	Adds regression tests for auth-failure classification and refresh response application (note: currently not wired into the repo’s default test runner).
.cspell/custom-dictionary-workspace.txt	Adds connack and emqx to suppress spellcheck noise in new/updated text.