On secondaries, some collections (replicated + oplog) should be read at the last applied timestamp whereas others (non-replicated) should be read at latest. In the shard role, we check all of the namespaces in the acquisition to see if any of them require last applied.

However, if we first do an acquisition for an unreplicated collection, we will use latest. And then if we do a later acquisition for a replicated collection, we will not check the read source as the snapshot is already allocated. This could lead to us reading from a replicated collection at latest on a secondary, which could result in reading mid batch application.