change karras scheduler sigmas to values recommended by SD from old 0.1 to 10 with an option to revert to old

fix 'RuntimeError: Expected all tensors to be on the same device' error preventing models from loading on lowvram/medvram.

make save to zip into its own button instead of a checkbox

make it so that memory/embeddings info is displayed in a separate UI element from generation parameters, and is preserved when you change the displayed infotext by clicking on gallery images

changed embedding accepted shape detection to use existing code and support the new alt-diffusion model, and reformatted messages a bit #6149

validate textual inversion embeddings

fix interrogate

set sd_model for API later, inside the lock, to prevent multiple requests with different models ending up with incorrect results #5877 #6012

fix(api): only begin/end state in lock

Add memory cache for VAE weights

fix rgba to rgb when using jpeg output

add additional memory states

remove unnecessary console message

fix shared state dictionary

Add CSS classes for the settings panels

Fix F541: f-string without any placeholders

Fix clip interrogate from the webui

implement train api

A recent change made the image RGBA, which makes the clip interrogator unhappy.
deepbooru and calling the interrogator from the api already do the conversion so this is the only place that needed it.