DOCS-12171 added clarification that delayed secondaries can ack write concern, but we should also include the recommendation that delayed secondaries should be non-voting and hidden.

A PSD replica set with a voting delayed secondary degraded to PD will be problematic if majority write or read concerns are used, because the majority commit point can only advance after replication to the delayed secondary.

I think the wording of "Should be hidden members" could also be stronger. This is closer to a Must outside of esoteric use cases.