potential slowdown when using run_iter over thousands of paths

[amyreese]

Walking paths with thousands of files, and then calling run_iter on those paths, seems to get slower and slower when pushing past ~10k paths, and eventually grinds to a near halt around 14/15k. Batching into groups of 1k paths eliminates the slowdown, so doesn't seem related to filesystem issues. Perhaps something in concurrent.futures.as_completed when dealing with that many futures.