[Feature request] Support "SVD 128/ExVid" Model?

### Feature Idea

Recently, ModelScope put out a SVD 128 model that claims to be able to render a total of 128 frames of video consistently, versus the default 25 for svd XT.

https://huggingface.co/ECNU-CILab/ExVideo-SVD-128f-v1/tree/main

However, their implementation uses a custom Unet, so the native SVD loader doesn't work with it.

(There may also be some issues with how it loads/processes input latents that aren't the default 512x512 resolution)

Humbly requesting that someone smarter than me can take a look at the implementation and add support for it to the current SVD nodes?

https://github.com/modelscope/DiffSynth-Studio/blob/main/diffsynth/pipelines/stable_video_diffusion.py

### Existing Solutions

_No response_

### Other

_No response_