Recently we have been having flaky job queue tests once again.

It might be good to spend a bit of time between main features about why we have these errors, and fix them properly. When I looked in the past, the problem was that those tests sometimes assume a very precise interleave between events, which we cannot always guarantee deterministically.