support vlm models for gguf format

[n1ck-guo]

gguf-py vlm support list and evaluation model (quant and infer):

• LlavaVisionModel: pixtral-12b, Mistral-Small-3.1-24B-Instruct-2503
• SmolVLMModel: SmolVLM-256M-Instruct
• Llama4VisionModel: Llama-4-Maverick-17B-128E-Instruct (packing)
• Qwen2VLVisionModel: Qwen2-VL-2B-Instruct, Qwen2.5-VL-7B-Instruct
• InternVisionModel
• Gemma3VisionModel: gemma-3-12b-it
• WhisperEncoderModel

[wenhuach21]

better list the vlm models we support or don't support at rtn mode if there are some gaps between ours and official's
better list the vlm models we support at tuning mode

[wenhuach21]

if possible, add CPU unit tests to verify basic functionality, and CUDA unit tests for more advanced scenarios, such as mixed-bit quantization, 8-bit for visual modules, and 4-bit for language module