fix(whatsapp): retry bridge spawn without CREATE_BREAKAWAY_FROM_JOB on access-denied

[Aperk357]

Summary

• plugins/platforms/whatsapp/adapter.py's subprocess.Popen call for the WhatsApp bridge used windows_detach_popen_kwargs() unconditionally, which sets CREATE_BREAKAWAY_FROM_JOB.
• When the parent Electron process's job object doesn't permit breakaway, CreateProcess fails with ERROR_ACCESS_DENIED, surfaced as PermissionError: [WinError 5] Access is denied — observed 1,271+ times in gateway.log, recurring every ~5 minutes.
• Every other detached-spawn call site already retries without the breakaway flag on OSError (canonical pattern: hermes_cli/gateway_windows.py::_spawn_detached). The WhatsApp bridge's Popen call was the one site missing this fallback.
• This PR mirrors _spawn_detached's exact pattern: wrap the Popen call in try/except OSError and retry with windows_detach_flags_without_breakaway() on Windows; re-raise unchanged on non-Windows (where there's no breakaway flag to strip).

Test plan

• Existing tests/gateway/test_whatsapp_connect.py suite passes unchanged (11 passed, 1 skipped pre-existing).
• Added TestBridgePopenBreakawayFallback with two new tests:
  • test_retries_without_breakaway_on_permission_error: mocks subprocess.Popen to raise PermissionError on the first call, verifies a second call is made with creationflags that exclude CREATE_BREAKAWAY_FROM_JOB (0x01000000), and that connect() succeeds using the retried process.
  • test_does_not_retry_on_non_windows: verifies the retry path is skipped on non-Windows and the original OSError behavior (single Popen call) is preserved.
• Full new suite run: python -m pytest tests/gateway/test_whatsapp_connect.py -q → 13 passed, 1 skipped.
• Not tested against the live running bridge (would risk interfering with production Hermes processes on this host).

🤖 Generated with Claude Code

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

fix(whatsapp): retry bridge spawn without CREATE_BREAKAWAY_FROM_JOB on access-denied

• tests/gateway/test_whatsapp_connect.py test_retries_without_breakaway_on_permission_error: assert "creationflags" in popen_calls[0] or True is a tautology — X or True always evaluates to True, so the first-attempt assertion can never fail. It should be assert "creationflags" in popen_calls[0].
• plugins/platforms/whatsapp/adapter.py connect(): the except OSError is broad — on Windows any Popen OSError (e.g. ENOENT if the node/bridge path is wrong) triggers a no-breakaway retry that will fail again for the same reason before propagating. Harmless (one wasted spawn), but narrowing to PermissionError (the documented WinError 5 case) would keep the fallback semantically precise and match the comment's intent.
• The retry re-specifies argv/env/stdout/stderr manually. Functionally equivalent today since windows_detach_popen_kwargs() returns only creationflags on Windows, but composing it as {**windows_detach_popen_kwargs(), "creationflags": windows_detach_flags_without_breakaway()} would keep a single source of truth for the spawn kwargs and stay correct if that helper ever grows more keys.

[alt-glitch]

This was generated by AI during triage.

Related to #65990: both repair the WhatsApp bridge breakaway failure on Windows. This PR preserves breakaway when available and retries without it on access denial; #65990 always omits breakaway. Maintainer selection is needed.