Adjust spawn_blocking() queue resize check

[Darksonn]

Let's avoid wasting work on allocating a larger buffer many times just because a single item was popped while the lock was not held.

cc @alex

[alex]

Seems was reasonable. Was this found with profiling a real workload or just reviewing the code?

[Darksonn]

Just reviewing the code.

[afurm]

The new condition new_queue.capacity() > queue.capacity() triggers a resize even when the queue isn't full. If the original queue had room but new_queue was overallocated (e.g., from a prior shrink_to_fit), this allocates a larger buffer unnecessarily. Is that trade-off intentional given the comment about not wasting the allocation?

[Darksonn]

I don't think we ever call shrink_to_fit() on the queue to begin with.