GitLab

Try to keep batch sizes more consistent which seems to improve things on
AMD GPUs.

Put them in models/clip

This should make things a bit cleaner.

It's pretty much the same as the LatentUpscale node for now but for images
in pixel space.

The models are modified in place before being used and unpatched after.
I think this is better than monkeypatching since it might make it easier
to use faster non pytorch unet inference in the future.