Memory not released by generate_stitch.py

[amcmahon01]

During long runs, generate_stitch.py appears to be holding objects in memory after they've been processed.

Possible "band-aid" solution: close and recreate multiprocessing pool for every day of processing. This wouldn't solve the root cause but would at least limit the accumulated objects. For reference, 5 days of processing has left most processes in the pool with ~5GB of memory usage (each). With a pool size of 20, this is basically utilizing all available memory.

[amcmahon01]

Solution for now:
Changed map to imap so that full lists aren't loaded into memory at once. Also added "maxtasksperchild" and "chunksize" parameters to help reduce overhead and release processes periodically. Generate_stitch and extract_features both create new pools for each day. There's still a serious memory hoarding issue, but it looks like it may now be originating in the cam objects.