Fixed RGB/BGR colorspace mismatch in GFPGAN implementation, added GFPGAN strength control to img2img/txt2img modes

Rearranged UI elements, added better auto-sizing, hid number arrows, made mode dropdown look like the generate button, removed unnecessary padding

fixed input verification not working properly with long textual inversion tokens in some cases (plus it will prevent incorrect outputs for forks that use the :::: prompt weighing method)
changed process_images to object class with same fields as args it was previously accepting
changed options system to make it possible to explicitly specify gradio objects with args

Workaround for gfpgan OOM on 6GB

force a gc if out of memory

fixed a bug with sampler selection (oops)

Added Real-ESRGAN to extras tab

settings tab

Support for using the script with textual inversion repo, should be safe for use in normal repositories too.

change the text output element to HTML

@hlky
torch_gc/empty cache after generation

commandline options for grid filetypes and max batch count

added torch_gc() which calls both cuda.empty_cache() and cuda.ipc_collect()
called before and after generation

Grid file named with seed and prompt, output as jpg

fix a bug in image resizing

added more info into readme

remove remote image as sample for img2img (it will now use a local image if you have one, none otherwise)

hide gradio progressbar for about 10% speed increase on my GPU; a commandline option to prevent hiding

added prompt verification: if it's too long, a warning is returned in the text field along with the part of prompt that has been truncated