Add Qwen2Audio Support

[arunasrivastava]

Hi!

I am working on benchmarking audio models for Qwen2Audio and would really love to see what features the model pays attention to for tasks like emotion recognition. would this be a feasible feature to add?