After SERVER-69573 a new oplog entry was added to handle persisted and in-memory state for the sharding's index catalog. As part of that work the in-memory state is changed in an observer. If the node is in a recovery mode, we already have a mechanism to restore the in-memory state to a consistent one, so, executing these operations in the observer seems unnecessary.

We could use the same mechanism as the recoverable critical section to prevent doing the in-memory state changes in recovery mode.