In case of a client directly connected to a shard of sharded deployment, write operations on orphaned documents are properly flagged in order to avoid notifications on change streams (see PM-2423).

The logic to identify if a write operation involves an orphaned document is relatively expensive, and it's currently triggered in replica set deployments as well. This is useless and wasteful as there are no orphaned documents.

The goal of this ticket is to review where the currently logic is used (e.g., here) and restrict it to nodes that are part of a sharded deployments.