also look into sorting contributions by payload size and tune `per_request` according to batches of similarly sized contributions