fix(qqbot): prevent busy-loop when WebSocket closes without exception

[ygd58]

Problem

When ws.closed becomes True, _read_events() exits its while loop cleanly without raising an exception. _listen_loop() immediately starts a new iteration with no sleep or reconnect delay, spinning at 100% CPU.

Fix

Two guards in _read_events():

1. Pre-check: raise if ws is already closed on entry
2. Post-check: raise after loop exits cleanly with ws.closed=True

Both surface as RuntimeError so _listen_loop schedules proper reconnect with exponential backoff.

Fixes #27810

[ygd58]

Checked #17704 and #20994:

• fix(qqbot): keep reconnecting after closed websocket #17704: adds pre-check only (if ws.closed raise on entry)
• fix(qqbot): prevent silent busy-loop when reconnect fails #20994: nulls _ws after reconnect failure

This PR (#27821) adds both pre-check AND post-check (raise after while loop exits cleanly with ws.closed=True), which is the missing piece in both. The post-check is what actually prevents the busy-loop in the 4009 session timeout scenario reported in #27810.

[BoardJames-Bot]

CI triage: the two failing checks are not from the qqbot WebSocket change itself.

• test fails only in tests/hermes_cli/test_aux_config.py because the PR's merged checkout is missing the current DEFAULT_CONFIG["auxiliary"]["session_search"] block while _AUX_TASKS references session_search.
• Scan PR for critical supply chain risks fires because the PR diff against current main includes hermes_cli/setup.py as an install-hook file (it appears as deleting the newer xAI OAuth setup changes from main). The workflow tried to post its detailed comment but fork PR tokens are read-only: Resource not accessible by integration (addComment).

Root cause: branch is stale relative to current main, not a qqbot regression. Next owner action: rebase/merge current origin/main into fix/qqbot-websocket-busy-loop and push the fork branch; that should remove the unrelated hermes_cli/setup.py/aux-config drift and re-run CI.

[BoardJames-Bot]

BoardJames CI triage: the test failure is the same merged-state aux-config drift I reproduced on current main, not caused by this PR's changes. tests/hermes_cli/test_aux_config.py expects auxiliary.session_search, but session_search no longer uses an auxiliary LLM and current config removed that block while leaving the picker/test stale. Opened #27835 to remove session_search from the auxiliary model picker and update the test; focused validation there: python -m pytest tests/hermes_cli/test_aux_config.py -q -o 'addopts=' -> 21 passed.

[teknium1]

Thanks for the QQBot reconnect fix. This specific busy-loop issue has since been fixed on current main by an automated hermes-sweeper review.

Evidence:

• gateway/platforms/qqbot/adapter.py:684 now raises RuntimeError when _read_events() is entered with a closed-but-non-None WebSocket, instead of returning normally forever.
• gateway/platforms/qqbot/adapter.py:639 routes that RuntimeError through the reconnect/backoff path in _listen_loop().
• tests/gateway/test_qqbot.py:2201 adds a regression test for the closed-but-non-None WebSocket case.
• The fix landed in 3eeca4613d618618093db416b564a2b9ef8dbe6a (fix(qqbot): stop 100% CPU spin when WebSocket is closed but not None (#31193, #31771) (#40574)).

Closing as implemented on main.