Fix race condition in trashtest_comms teardown

[Badg]

Unix systems appear to be having a race condition in server teardown in trashtest_comms. Currently, the trashy workaround is a time.sleep(.1) call. This is bad practice and slows down the tests. It would be nice to know where the race condition is coming from, and then to fix it (theoretically, stop_threadsafe should block on full close. Is that the same as releasing the socket though?). It's possible this is a problem with Loopa.