[Feature] Add Support to Video Generation Models and architectures

### Motivation

## 🚀 Feature Request: Add Support for Video Generation Models

Please consider adding support for the following **video generation** models, which are currently not supported by `lmdeploy`:

- [CogVideoX-5b by THUDM](https://huggingface.co/THUDM/CogVideoX-5b)
- [Mochi-1 (Preview) by Genmo](https://huggingface.co/genmo/mochi-1-preview)
- [Allegro by Rhymes AI](https://huggingface.co/rhymes-ai/Allegro)
- [LTX-Video by Lightricks](https://huggingface.co/Lightricks/LTX-Video)

These models represent the latest advancements in **text-to-video** generation, and their support would significantly enhance lmdeploy's capabilities in the multimodal AI space.

## 💡 Additional Context

These models are gaining popularity in generative AI workflows for creating high-quality video content from text prompts. Integration with lmdeploy would enable more efficient inference and unlock a range of creative and industrial use cases.

### Related resources

_No response_

### Additional context

_No response_