Fix race condition in Connection.serve and AsyncResult.wait

[notEvil]

#530

The first commit adds the unit test, the second contains the fix. Checkout d861ab9 to confirm the issue using

python -m pytest ./tests/test_race.py

[comrumino]

bcab6a0 has some refactoring that may interest you. by no means final, but another pass.

[notEvil]

It completely breaks thread binding, and introduces a busy loop here bcab6a0#diff-506bd4e029401c85a178488d45b05378da956b6f608c64a6e5f9aba74890e8c1R56. I wouldn't change the way thread binding works unless there is a clear gain because there is lots of potential to introduce race conditions (its complexity opens up a lot more windows than default serve). To be fair, the implementation looks bad and overly complicated. There is potential to make it more readable, but the base design is conceptually sound imo. (#491 (comment))

• replaced _receiving with _recvlock._is_owned as it is slightly more
  robust

how is _receiving not robust?

[comrumino]

While I broke some things w/ my changes, there were different things were broken prior to it. Test results about latest comment in #530