there was a chance that by the time the assertion is executed,
the replica already manages to reconnect.
now we make sure the replica is unable to re-connect to the master.
additionally, we wait for some gossip from the disconnected replica,
to see that it doesn't mess things up.
unrelated: fix a typo when trying to exhaust the backlog, one that
didn't have any harmful implications
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
d04f306931