When running:

coll.aggregate([{$group: {_id: "$a", accum: {$first: "$b"}}}])

on a collection that contains indexes: {a_1, b_1, a_1_b_1}, the multiplanner always picks a scan on `a_1` + FETCH. 

The multiplanner does not even consider using a_1_b_1, which would avoid the FETCH entirely. 

It seems that we only consider the distinct field here when trying to pick the index with the fewest fields. Instead, we should also account for any projections in that function.

Add a corresponding test in distinct_aggregation_multiplanning_md.js