tests: correct the output order about starting a background thread for Windows
I didn't track down where this change occurred. I assume it's related to some
buffering changes, and/or an explicit flush somewhere.
Differential Revision: https://phab.mercurial-scm.org/D9539