Still should probably rename wait_writable

Should (hopefully) fix travis builds

a java echo server! a pyopenssl equivalent that somehow seems to be
working now? the body of _retry is duplicated 3 times so I can get
more detailed debugging and coverage information depending on who
called it!

Maybe a bit fragile though... checkpointing before attempting a new idea

And use it in MockClock to get better behavior when using
wait_all_tasks_blocked and autojump_threshold=0 together. Potentially
useful in general as well.

This is starting to look like something real!

And ssl.py is approaching a 2:1 ratio of comments to code, which is
probably about right.

(it's an AbstractAsyncContextManager, but that doesn't exist, even in
py36)