option to keep multiple models in memory

[AUTOMATIC1111]

Description

• option to specify how many loaded models you want to keep in memory
• additional checkbox that lets you decide whether you want to keep just the one model in VRAM or all of them
• if switching to an already loaded model, it's either instantaneous (if the checkbox is not checked) or very fast, something like 1s for me (if the checkbox is checked)
• obsoletes the Checkpoints to cache in RAM setting.
• tested to work with --medvram
• Loras tested to work properly
• additional things I put into the PR despite telling others to not put extra stuff
  • suppressed sgm/ldm print statements can be restored in settings
  • do_inpainting_hijack removed, the function is hijacked at startup instead
  • fixed a bug where get_empty_cond would create empty prompt cond using Lora weights

Screenshots/videos:

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

[ghost]

This introduced an issue where models get corrupted after reusing the older model to load the new.
Read #13516 for a temporary fix until it's sorted out.