[Feature] Support for CogVLM2-Video-LLama3-Chat in TorchEngine

### Motivation

Hi team,

ZhipuAI just released their multi-modal model `CogVLM2-Video-LLama3-Chat`. Can we support its serving with TorchEngine? It seems that they use a new causal model architecture called `CogVLMVideoForCausalLM`.

I’m really excited to try out this model with lmdeploy.

Here's the link for reference: [CogVLM2-Video-LLama3-Chat Model](https://modelscope.cn/models/ZhipuAI/cogvlm2-video-llama3-chat/file/view/master?fileName=modeling_cogvlm.py&status=1).

Thanks!

### Related resources

_No response_

### Additional context

_No response_