Upscaler model loading cleanup

[akx]

A summary of changes in code

• drop the dependency on basicsr for downloading stuff, make its arguments explicit (it just wrapped torch.hub anyhow)
• clean up some pretty suspicious URL handling (and mark a couple TODOs for a bit later)
• use exceptions, not "lol return None" for error handling

"a bit later" refers to some future work to clean up some of the upscaler stuff which is a bit baroque atm.

Tested:

• ScuNET upscaling in HR fix
• SwinIR upscaling in HR fix
• Real-ESRGAN upscaling in HR fix
• LDSR upscaling in HR fix

Checklist:

• I have read contributing wiki page
• I have performed a self-review of my own code
• My code follows the style guidelines
• My code passes tests