This appears to be something of a magnet for accidentally writing flakey simulation test cases.
This showed up in PR #13252 and independently, somebody mentioned this failure mode to me offline a few months ago. I suspect it's a recurring thing.
* deleted expectedRate from Cycle test, as it is a magnet for rare failures in typical cases where Cycle is combined
with error injection. Also, Cycle need not concern itself with detecting performance regressions.
* add a MinimumThroughput workload to make up for potential loss of coverage elsewhere
* fix test bug on rate calcuation and omission regarding injected failure suppression
* fix it some more
* move comment to avoid AI complaining about not mixing it with the apache copyright
gxglass