We should look into allowing multiplanning to decide between index scans (if any are available) and distinct scans instead of a collection scan in a case where we have distinct("y") and an index like {x: 1, y: 1}