Worsening of selftest thread leaking

[Robert Collins]

So, I put some comments in a bug about debugging this.

However, we essentially need to make sure that all started threads are
joined properly. *I* would like to have an initial hook added (e.g. as
the first line of setUp) that there are no different threads running
when a test finishes than when it started.

That, it seems to me, would squash this pdq.

-Rob