Supporting `*.safetensors` format.

Remove cmd args requirement for deepdanbooru

Added a flag to run as root if needed

fix the model name error of Real-ESRGAN in the opts default value

add model_name pattern for saving

no-half support for SD 2.0

Fixed issue with selected_gallery_index()

#3449 - VRAM leak when switching to/from inpainting checkpoint

Fix setting device for GFPGAN and CodeFormer

Fixes for PyTorch 1.12.1 when using MPS

torch.cuda.empty_cache() defaults to cuda:0 device unless explicitly …

Add DPM++ SDE sampler

add an option to choose a different temporary directory in the UI
add an option to cleanup the selected temporary directory at startup

torch.cuda.empty_cache() defaults to cuda:0 device unless explicitly set otherwise first. Updating torch_gc() to use the device set by --device-id if specified to avoid OOM edge cases on multi-GPU systems.

Probably redundant, considering the model name *is* a filename, but I suppose better safe than sorry.

If a model file exists with extension `.safetensors` then we can load it
more safely than with PyTorch weights.