Right now the GPU IDC always uploads the full image (possibly scaled). In cases where a small sub-rect of an image is needed, the GPU IDC could avoid  uploading as much data by just uploading the subrect. This would use a similar approach to scales, where we would need to re-upload a larger subrect if more of the image was needed.

This would also let us use the scaling optimizations with sub-rects. Right now (due to bleeding), we can't scale an image which has a sub-rect extracted.