[Feature]: Load/Evaluate `W4A16` Qwen/Qwen-Image on vllm-omni


### Feature Description

As the tile

### Motivation and Use Case

Model: https://huggingface.co/Qwen/Qwen-Image
Target dtypes: w4a16

### Alternatives Considered

_No response_

### Definition of Done

- [ ] Load the quantized model
- [ ] Align the accuray w/ the qdq mode.

### Additional Context

_No response_