reduce `per_request` dynamically in the client if maximum payload size reached

[tschaume]

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