Support loading original stable diffusion checkpoints

[rnxyfvls]

• Add checkpoint_loader based on code from https://github.com/kjsman/stable-diffusion-pytorch/issues/7#issuecomment-1426839447
• Add example program

[kjsman]

Thank you for your contribution! Some thoughts:

• It seems these patches have largely three points:
  • Converting the original SD checkpoint to the model which is compatible to this repository
  • Converting the safetensor model to the compatible model using the first component
  • The minimal demo Python code which "just works" with the given prompt, without using Jupyter Notebook.
• I like all of them and want to ship them all, but how about separate them and name them clearer? I'm thinking of creating a stable_diffusion_pytorch/convert_from_sdmodel.py (which does same thing with checkpoint_loader.py), safetensor/convert_model.py (convert the safetensor model and saves to somewhere), and safetensor/generate_image.py (convert the safetensor model on-demand and generate image; do the same thing with demo_loadmodel)
• I think using the official Stable Diffusion v1.5 safetensor model would be better for this general demo. I also like custom models such as Anything :) but I also want to keep things more official and academical in this repository.
• The checkpoint converting code is written by @treeform I guess? We would need their approval to distribute this code as MIT license.

[rnxyfvls]

• Move conversion code to stable_diffusion_pytorch/convert_from_sdmodel.py: done
• Change demo to stock SD 1.5 base model: done
• convert_model.py: done, I have put it in the root directory though as it needs imports from stable_diffusion_pytorch/, and main programs can't import from parent directory
• Demo program demo_sdmodel_generate_image.py: also in root directory for the above reason
• Code license: I have asked treeform on issue #7 , if I can't get a response from him I will rewrite that part from scratch