Add tests for aborted Pool::acquire and Semaphore::acquire calls showing a memory leak

[bikeshedder]

I wanted to show the reason for the deadlock I reported in #2 but found a memory leak instead.

A aborted Semaphore::acquire does leave the waker behind in the queue which causes the memory to grow indefinitely.

The thing I don't understand: I wanted to show that the aborted task A leaves a waker for itself in the queue and when the permit is returned task A is woken instead of task B thus causing the code to deadlock. This does not happen. Instead I found that calling waker.wake() for task A actually wakes task B.

I'm trying to wrap my head around why it works that way and doesn't deadlock. Whatever the reason for that behavior is. It's a memory leak nonetheless.

[codecov-commenter]

Codecov Report

Merging #3 (26322c6) into main (9702306) will increase coverage by 4.96%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main       #3      +/-   ##
==========================================
+ Coverage   85.14%   90.11%   +4.96%     
==========================================
  Files           6        6              
  Lines         202      263      +61     
==========================================
+ Hits          172      237      +65     
+ Misses         30       26       -4     

Impacted Files	Coverage Δ
qp/src/pool.rs	84.54% <100.00%> (+5.02%)	⬆️
qp/src/sync.rs	92.68% <100.00%> (+13.51%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c208a2c...26322c6. Read the comment docs.

[Astro36]

The thing I don't understand: I wanted to show that the aborted task A leaves a waker for itself in the queue and when the permit is returned task A is woken instead of task B thus causing the code to deadlock. This does not happen. Instead I found that calling waker.wake() for task A actually wakes task B.

I don't know why this happens, but the problem that the queue is not completely empty seems to be caused by calling the Acquire::poll function once more when the Acquire future is cancelled.

This is a commit that just fixed the memory leak issue.

Thank you for your enthusiasm!