fix: resolve entity unavailability, stale tokens, and unresponsive controls

[hugo-brito]

Summary

Four code fixes for V02 (AWS IoT) device support, plus repo maintenance fixes for tests and CI. Addresses persistent entity unavailability, authentication failures, sluggish dashboard controls, and log spam from missing keys.

Files changed: 8 | +365 -34 | 78 tests passing

Code fixes

1. Ignore unreliable is_online flag (entity.py)

The Bestway/Gizwits cloud API frequently reports is_online: false even when the device is functioning normally and controllable via the official app. The API continues returning valid device state data regardless of this flag.

The base BestwayEntity.available property gated all entities on self.bestway_device.is_online, causing all spa entities to become permanently unavailable. The connectivity binary sensor continues to report the raw is_online value as a diagnostic indicator.

Fixes #89, #93, #100

2. Always re-authenticate on startup for AWS IoT (__init__.py)

Stored tokens expire server-side without any expiry field the integration can check. When a stale token existed, the integration skipped authentication and proceeded to refresh_bindings(), which failed with "Token is not authorized", leaving all entities unavailable until the next HA restart.

Now always requests a fresh token on startup. Re-authenticating is a single POST request (cheap).

Related: #86

3. Optimistic state updates for responsive controls (switch.py, climate.py, select.py)

Switch toggles waited for a full API round-trip (2-10s) before reflecting the state change in the UI. If the post-command refresh failed (e.g., temporary API timeout), the UI reverted the toggle, requiring the user to toggle twice.

• Switches now use optimistic state tracking - the UI updates instantly, then a non-blocking background refresh confirms the actual state.
• Climate and select entities now use async_request_refresh() (non-blocking, debounced) instead of async_refresh() (blocking).
• Select (bubbles) was missing a refresh call entirely.

4. Safe attribute access for Tunit key (climate.py)

When the API returns a partial state (e.g., after a timeout or during startup before the first full poll), the Tunit key may be missing from attrs. Using dict indexing caused a KeyError that spammed the logs (~200 occurrences over 4 days). Changed to .get("Tunit", 1) to safely default to Celsius (matching the no-status fallback behavior).

Repo maintenance fixes

5. Defensive async_unload_entry (__init__.py)

Use .get(DOMAIN, {}).pop(entry.entry_id, None) to handle the case where hass.data[DOMAIN] was never populated (e.g., failed setup). Prevents KeyError during unload of partially-initialized entries.

6. Config flow test teardown (test_config_flow.py)

• Patched both async_setup_entry and async_unload_entry in the bypass_setup_fixture to prevent teardown KeyError when HA manages entries that were never fully initialized.
• Added verify_cleanup fixture override that joins daemon threads before the upstream thread assertion runs. The upstream verify_cleanup races with _run_safe_shutdown_loop from asyncio's shutdown_default_executor().

7. CI workflow fixes (.github/workflows/test.yaml)

Fixed two copy-paste bugs from the givenergy_local project:

• branches: [master] -> branches: [main] (push-triggered CI never ran)
• --cov custom_components.givenergy_local -> --cov custom_components.bestway (coverage measured wrong package)

Testing

Tested on a V02 Airjet spa (UltraFit, product_id T53NN8) over 5 days (April 5-10):

• All 10 entities remain available across multiple HA restarts
• Controls (power, filter, jets, bubbles, heater, temperature) respond instantly
• Zero "Token is not authorized" errors over the monitoring period
• Connectivity binary sensor correctly reports raw is_online value
• Tunit KeyError log spam eliminated

Test suite: 78 passed, 0 errors

11 regression tests in test_availability_and_controls.py exercising the actual entity classes:

• Entity availability with is_online=True, is_online=False, missing device, coordinator failure
• Switch optimistic state: _assumed_state flag, optimistic turn on, state cleared on coordinator update
• Climate Tunit safety: Tunit present (Celsius), Tunit=0 (Fahrenheit), Tunit missing (defaults to Celsius), no status (Celsius)

Compatibility

• V01 (Gizwits) configurations continue to function - available property change applies to the shared base class, and climate/switch changes use the same patterns for both backends.
• No migration required.

[hugo-brito]

@cdpuk, could you please review?

[cdpuk]

Thanks for the collection of fixes - I don't have a V2 device so very much appreciate community contributions. I'll give this a quick check over the next few days. Could you just take a look at why the pre-commit checks are failing?

[hugo-brito]

@cdpuk no worries. I added a few more fixes and updated the PR description. The precommit check is now passing.