Currently the $group tracks memory so it knows when it should spill. The memory usage tracker for this lives in GroupProcessor. We should update the group stage so that it checks a feature flag, and if it's true, updates the memory tracker with the upstream operations-scoped tracker, which will report memory for the stage into CurOp.

With this in place, we can write end-to-end JS tests that can show that profiles and slow query logs will show the metrics.