Initial round of Music Separation improvements / feature adds

[RyanMetcalfeInt8]

• Music Separation can now support the following models:
  • htdemucs v4 (the one we've used all along)
  • htdemuvs v4 fine-tuned drums (single stem)
  • htdemuvs v4 fine-tuned bass (single stem)
  • htdemuvs v4 fine-tuned others (single stem)
  • htdemuvs v4 fine-tuned vocals (single stem)
  • htdemuvcs v4 6s (6 stems, includes guitar, piano)
  • MelBandRoformer vocals (Kimberley Jenson)
  • (and easily other MelBandRoformer models, to add later)

The implementation for htdemucs was replaced by one that adheres to a new 'demix' interface, which is common between HTDemucs / MelBandRoformer models.

Music Separation GUI enhancements to select the new models, and to change the drop-down 'mode' selection based on the capabilities of the current selected model.

Our huggingface repo's still need to be updated with these new models. Will submit a new PR with those changes once they are populated.