Reserve quota before asynchronous preparation

Concurrency quotas must reserve capacity before asynchronous preparation that can later launch capped work. Counting only when the expensive call starts is unsafe: a fast task can finish and free the in-flight pool while slower preparations are still pending, allowing replacements that later cross the boundary and exceed the quota. Convert each reservation into consumed budget immediately before the external call; release it in finally only when that call never started. Test mixed latency, not only equal-duration overlap: hold one preparation, complete another, and prove no replacement enters until a reservation is released.

Applies when: A rolling worker pool has asynchronous eligibility, locking, or setup before bounded external work. Keywords: concurrency quota reservation async preparation worker pool oversubscription mixed latency