fix(daemon_pool): support Python 3.14 WorkerContext API

[jcrabapple]

CPython 3.14 removed ThreadPoolExecutor._initializer / _initargs and replaced them with a WorkerContext pattern, and _worker() changed from 5 args to 3. This breaks DaemonThreadPoolExecutor on any system running Python 3.14 (Fedora 44, etc.) with:

AttributeError: 'DaemonThreadPoolExecutor' object has no attribute '_initializer'

This manifests in the CLI as intermittent tool-call failures during parallel operations (skill_view, web_search, search_files, and any other concurrent tool dispatch).

The fix detects the available API at runtime via hasattr(self, '_initializer') and branches accordingly — old path for 3.8–3.13, new WorkerContext path for 3.14+. Verified on both Python 3.11 and 3.14.

The daemon pool never passes a custom initializer, so None/() is the correct equivalent in the 3.14 path.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #65182. Both patch tools/daemon_pool.py for the same CPython 3.14 WorkerContext/_worker compatibility change; #65182 is the earlier open implementation.

[spfcraze]

Independent verification of this fix (Python 3.14.4, Linux x86-64, 2026-08-10).

I independently hit the same crash — AttributeError: 'DaemonThreadPoolExecutor' object has no attribute '_initializer' on 3.14 — and implemented a near-identical WorkerContext fix (hasattr-guarded branch). Since our approaches converge on the same mechanism, here's the sabotage-verified evidence I generated while validating:

Scenario	Result
test_daemon_pool.py on 3.14, unfixed tree	5/5 FAIL — AttributeError: _initializer
Same tests on 3.14 with the WorkerContext fix	5/5 PASS
Same tests on 3.12 with the fix	5/5 PASS (backward compatible)
test_delegate.py on 3.14, unfixed	41 failed / 64 passed
test_delegate.py on 3.14, fixed	63/63 PASS
test_compression_concurrent_fork.py + test_async_delegation.py, fixed	62/62 PASS

Key detail worth noting in the fix: the crash is NOT only in daemon_pool.py's own tests — delegate_tool.py:2802 uses this pool for its subagent timeout executor, so the fix unblocks delegation/orchestration on 3.14, not just the pool itself. My regression tests run the pool with an initializer/initargs in a subprocess (the exact 3.14 WorkerContext path) and verify workers receive the initialized thread-local.

No changes requested to this PR — just confirming the fix is correct and load-bearing from an independent run. Happy to add the regression tests here if useful.