nmlc chokes (then eventually recovers) on some images

[gravitystorm]

nmlc v0.3.1 chokes on a small number of images, and can take 15-45 minutes to process them. It's totally unclear as to why and what goes on, but they do eventually finish.

It can be seen by running watch lsof -c nlmc during compile - most images are processed quickly, but some stick around. I tried swapping out the images themselves to no avail so it's unlikely to be the contents of the image, perhaps the number if images processed so far, or some other resource exhaustion.

The plan will be to try with a python3 (0.4.0) version of nmlc, and if that still shows the same thing, start chopping up the nml files to create a testcase.