v0.3.0

v0.3.0

Fix thread pool shutdown hangs and dropped work in the zero-fast-queue path.

Concrete changes:

• treat fast_queue_capacity=0 as a true fast-queue disable and route tasks directly to the slow queue
• track pending tasks explicitly and let join() wait for the pool to drain before stopping workers
• serialize task acceptance and worker cleanup with a lifecycle mutex to reduce submission/cleanup races
• keep workers draining queued work during STOPPING and make futured_post fail fast with an exception after shutdown
• add regression coverage for repeated join+future stress and post-after-join behavior