CPU bottleneck during network training

[aleide]

The generation of training data takes up a disproportionate amount of time during training.

[aleide]

This is being worked on in branch: al-multiprocess-imgen

[hsaga]

I deleted this branch as it had not been updated (and couldn't be at this time, if I remember correctly).