They're not very good tests, but they do illustrate the problems with
the current implementation of the futures Sink/Stream.

The implementation is pretty crappy! It's full of TODO comments. I
didn't reorganize the code in a way that preserves much of the original
semantic clarity, which was definitely the first mistake. It works, and
does demonstrate one way of factoring around sometimes blocking and
sometimes not.

The tests might not be comprehensive enough either. I'm not sure.