We expect to resume from the last valid-as-of timestamp or last scanned oplog entry when tailing the oplog and tassert if the previously processed entry is lost, which suggests a potential data loss and durability issue for the size and count metadata.

However, in production this will loop infinitely and block all future flushes. If not manually remediated in time, this could become an availability issue as well. 

We should keep the strong assertion 

• To crash in testing
• To assert in production and trigger AF generation
• But also allow the tailer to continue so we don't keep increasing backpressure from later oplog entries, which could make the recovery harder