On mongos, the effective read/write concern (client-supplied merged with the cluster-wide defaults is computed in the Service Entry Point and stored on the opCtx. But applying it onto the command forwarded to the shards is left to each command implementer, who must remember to call setReadWriteConcern()/applyReadWriteConcern(), which is error-prone.

Though SERVER-127620 made this application more consistent, the next step is to move the merging into one single place on the router ServiceEntryPoint
so command implementations no longer need to call setReadWriteConcern() themselves. The per-command calls can then be removed.