the last PR broke saving EXiF completely for me. I don't know if it was broken already or some condition changed, but it seems like the person who originally added EXIF said, saving it with PIL may not work. I switched to using piexif to add data after the file written.

* image.save takes exif as a parameter
* piexif takes the bytes as a parameter, not the exif_bytes function itself
* reduce calls to create_exif_bytes

As it is currently, txt2img and img2img send back the previous output args (txt2img_gallery, generation_info, html_info) whenever you generate a new image. This can lead to uploading a huge gallery of previously generated images, which leads to an unnecessary delay between submitting and beginning to generate.

prevent replacing torch_randn globally (instead replacing k_diffusion.sampling.torch) and add a setting to disable this all

 Prompt editing only applies to images in first batch of desired batch size when batch count > 1 #535

fix loopback color correction to store color_correction info for initial image and use that instead of extracting from previous image in a loop #481 #541

... instead of removing them altogether.
This makes the prompt editing filenames [old:new:step] easier to read

fixes bug #464, which happens when controls is None, which leads to args_to being None.

Setting empty envars causes the `os.environ.get('ENVAR', "some default")` pattern in launch.py to ignore it's fallback value.

* fix for odd error only in webp files whereby piexif.insert inserts the bytes correctly, but image.save inserts extra "Exif" in the image metadata which results in an error on reading