[pa-devbot]

Fixes ThreadPool_Default::dispatch() never growing the thread pool.

The bug

dispatch() enqueued the task and called m_thread_cv.notify_one(), but never called spawn_threads(). Since all of the global pools in GlobalThreadPools.cpp start with zero threads, a plain dispatch() on a cold pool had no waiter to wake - the task sat in the queue and never ran. On a pool with fewer live threads than its maximum, the pool also never used its headroom.

dispatch_now_blocking(), try_dispatch_now() and run_in_parallel() all call spawn_threads(), so only the plain dispatch() path was affected.

The fix

Call spawn_threads() after the task is enqueued (under the lock), the same as the other entry points.

Testing

Standalone functional test against ThreadPool_Default:

• a plain dispatch() on a 0-thread pool now runs (previously it hung forever),
• 4 concurrent 300 ms tasks on a 0-thread pool (max 4) overlap - 301 ms wall time, peak concurrency 4 (previously serialized),
• 8 tasks on a max-4 pool stay capped at 4 concurrent and run as two waves,
• dispatch_now_blocking() / try_dispatch_now() still work.

pa-build + pa-test green.