Enhancement: Implement DistVAE

[cb88]

It would be nice if multi GPU patching of VAEs could be integrated in cpp in addition to the improvements to tiling being made. It would allow speeding up VAE as well as reducing VAE memory spike also for large images (tiling also does this as well). Note the input patching would decrease memory spike in all scenarios tiling or not.

Xdit is also licensed MIT.

Implementation:
https://github.com/xdit-project/DistVAE

References:
https://github.com/xdit-project/xDiT/blob/main/docs/methods/parallel_vae.md
https://hanlab.mit.edu/blog/patch-conv