In HELP-99223, we could see that replication lag was building, but could not understand why that was (see also SERVER-134905). We should ensure that metrics are sufficient to do a reasonably confident RCA based on FTDC data.

Specifically, we should be able to differentiate between the fetcher, writer, and applier being the bottleneck. We should also be able to understand how many threads are active in each of these, and what they are spending their time doing (e.g. blocked by backpressure, blocked in storage, idle).