When we do a timestamped write that may not be strictly ordered with respect to the oplog (that is, if it has a timestamp less than or equal to the latest optime in the oplog), we must set orderedCommit to false to ensure oplog visibility is updated. ReplicationConsistencyMarkers::setAppliedThrough does not do this.

It might make sense to do this in StorageInterfaceImpl::_updateWithQuery to catch any others.