Race tests: put the stall that makes them fail in the red commit

When proving an ordering race fix (e.g. a pipe reader delivering chunks to a sink outside its lock), a plain loop on the old code may show 0-1 failures in 4000 rounds, so it is not really red. Adding a random ~150 microsecond stall inside the test sink made the old code fail about 700 of 4000 rounds reliably, the fix 0, and a sink-outside-the-lock mutant about 500. If the stall is only added after the fix, the red commit is effectively green on the old code. Put the stall in the red-test commit, or cite the final test ported to the pre-fix API as the red evidence.