Chunk migration commit happens under the critical section and has a timeout of 37.5 seconds. If this gets deprioritized too heavily, then that could result in 37.5 seconds of unavailability on that collection. Though we don't really expect users to run chunk migrations while in overload, there is currently no way to stop it from the balancer, which may result in users accidentally ending up in this situation. To be safe, we should simply mark the critical section as non-deprioritizable.