This change ensures that, after sending data to one thread, fuzz_hook waits
until the data has been written out by the other (or until a short timeout
has occurred.) This prevents race conditions where the second packet is
processed by e.g the display-side loop before it receives the first from the
application-side loop, making runs of fuzz_hook unstable and more difficult
to reproduce. As most round trips require several context switches, and some
packets are dropped entirely, this change significantly slows down test runs.