Handling autoCompact in stopReplicaSetWrites requires a lot of complexity, we need to remember the current state, disable it with retries (because WT can fail), handle rollbacks and reconciliation, consider the case where autoCompact is re-set with bypass, etc.

autoCompact should increase disk usage very slightly and it's probably not worth supporting it. Analyze the real usage on the fleet by looking at metrics, how often we stop writes on clusters that use it, and consider removing all complexity related to it.