My particular angle of motivation on the refactor is from SERVER-67424 wherein I encountered the following problem:

"It appears that we modify read concern settings (prepared conflict) in DocumentSourceWriter. The problem with this is that if the high level code has a AutoGetLockFree*, we've already got a storage snapshot open: read concern is normally manipulated in the AutoGet* or earlier layers. It works today because we happen not to hold AutoGet* across runAggregate's use of DocumentSourceWriter. I was trying to hold locks across this, to hold a lock across registration of CursorManager::registerCursor for concurrency control with replication rollback."

Locking uncertainty in runAggregate hinders development: it's very confusing.