fix(daemon_pool): support Python 3.14 ThreadPoolExecutor WorkerContext signature

[xxiaoxiong]

Problem

Python 3.14 refactored concurrent.futures.ThreadPoolExecutor: the _worker target now takes (executor_ref, ctx, work_queue) where ctx is a WorkerContext produced by self._create_worker_context(), replacing the legacy (executor_ref, work_queue, initializer, initargs) signature. __init__ also no longer sets self._initializer / self._initargs.

DaemonThreadPoolExecutor._adjust_thread_count still passed the legacy 4-arg shape, so on Python 3.14 every spawn raised:

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

This breaks every daemon-pool consumer on 3.14+:

• tools/delegate_tool.py (subagent batch execution + timeout wrappers)
• tools/async_delegation.py (background delegate_task dispatch)
• tools/skills_hub.py (catalog fan-out)

When the daemon pool cannot spawn, delegate_task's batch path collapses to a synchronous inline run, destroying parallelism and silently wedging subagent fan-out.

Fixes #63769.

Fix

Detect the new layout once at import time by inspecting _worker's signature (param count 3 = 3.14+, 4 = 3.8–3.13). The _worker signature is the load-bearing contract: stable across patch releases and immune to the instance-vs-class attribute pitfall that breaks hasattr(ThreadPoolExecutor, '_initializer') (3.11 only sets it on instances, so the class check is False on every version).

The 3.14+ branch passes self._create_worker_context() as ctx; the 3.8–3.13 branch is unchanged. Daemon behavior (daemon=True, no _threads_queues registration) is preserved on both paths.

Testing

tests/tools/test_daemon_pool.py:

• Existing 4 tests still pass on 3.8–3.13 (unchanged behavior)
• New: test_python_314_worker_context_signature_does_not_crash — guards the regression: submit must not raise AttributeError, worker must be daemon, must not be in _threads_queues, initializer must run via WorkerContext. Skips on 3.11–3.13.
• New: test_python_314_submit_many_tasks_exercises_reused_workers — guards against worker-context-tied-to-first-submit regression on the idle-semaphore reuse path. Runs on all versions.

Verified on both runtimes

Python	daemon_pool tests	delegate + async_delegation tests
3.11.15	5 passed, 1 skipped	27 passed (no regressions)
3.14.4	6 passed (regression test now executes)	—

Notes

• The detector uses _worker's function signatur e via inspect.signature, not class-level attribute probes — this is the only reliable cross-version check.
• _PY314_PLUS is a module-level constant computed once at import time, so the hot path (_adjust_thread_count) stays branch-lite.
• Test helper _supports_worker_context() re-uses _PY314_PLUS so the gate and dispatch logic stay in lockstep.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #57459 (earliest open) — same tools/daemon_pool.py _adjust_thread_count site and same Python 3.14 _worker-signature crash. Mechanism differs only in the detection probe (inspect.signature here vs getattr fallback in #57459). Issue #58596 is the bug spec; #47634/#50077 patch the distinct async_delegation.py site (related).

[xxiaoxiong]

Closing — Duplicate of #57459 (Python 3.14 daemon_pool compat, same site/issue, different probe mechanism)