Add support of Qwen2vl

**Is your feature request related to a problem? Please describe.**
Currently I am using Qwen2vl, this is the best vlm model for my project. I hope llama-cpp-python can support this model. I tried to use llama.cpp to build a server, but llama.cpp is not allow to use mm-proj.

**Describe the solution you'd like**
Support the qwen2vl model and use it like other vlm models. 


