fix(qqbot): stop 100% CPU spin when WebSocket is closed but not None (#31193, #31771)

[teknium1]

Summary

Fixes a 100% CPU busy-loop in the QQ bot adapter when the WebSocket is closed but the object isn't None. Closes #31193 and #31771.

_read_events() returned normally in that state (its while condition is false on entry); _listen_loop treats a normal return as a clean read, resets backoff to 0, and immediately retries — pinning a core. Raising on entry routes it through the proper reconnect/backoff path.

Changes

• gateway/platforms/qqbot/adapter.py: raise RuntimeError when self._ws.closed on entry to _read_events.
• tests/gateway/test_qqbot.py: regression tests for closed-ws and None-ws entry.

Validation

TestReadEventsClosedWsGuard 2 passed.

Salvaged from #40497 (@xushibo) and the byte-identical #40291 (@cnfi); both credited.

[github-actions]

🔎 Lint report: salvage/40497-qqbot-spin vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9962 on HEAD, 9962 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5167 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[mohamedorigami-jpg]

Clean and minimal fix for the busy-loop. The guard correctly raises RuntimeError instead of letting the while condition short-circuit to False on a closed-but-non-None ws. However, the test for ws is None also raises RuntimeError via the same path -- but the existing pre-check if not self._ws: raise RuntimeError("WebSocket not connected") on line ~683 already handles that case. The new self._ws.closed check runs after that pre-check, so for a None ws, we never reach the new code (the not self._ws guard raises first). The test_read_events_raises_when_ws_none test expects the same RuntimeError but is actually testing the pre-existing guard, not the new one. Consider adjusting the ws=None test to verify via the specific error message ("WebSocket not connected" vs "WebSocket closed") so the regression coverage is unambiguous, or remove it if the pre-existing guard is already tested elsewhere.