One would think that the CappedInsertNotifier is used to notify about new data in a capped collection. However, the replication system hijacks this notifier for the oplog to get the query layer to generate empty getMore batches and propagate the majority commit point.

Since this is effectively just a replication component, we should pull this component out of the storage engine, and have replication build a specialized component for the oplog query layer.

Idea:

• Have an OplogNotifier that is owned by Replication
• When the allDurable timestamp is advanced, Replication notifies the OplogNotifier
• When the majority commit timestamp is advances, Replication notifies the OplogNotifier

In this way, the storage engine is completely uninvolved.