We should consider trying to bypass some of the work to materialize to BSON then instantiate a Document when going from a covered query in find to an aggregation that consumes the output of this query via a DocumentSourceCursor. Some flame graphs from SERVER-87237 suggested we spend quite a bit of time per document in this code, but in the case of covered queries, we can access alot of the information we need directly from the index and instantiate a Document directly (which could save us time if we need to do Document-friendly things later on in the pipeline).