[WIP] [STEP 2] split compressor into few quantizers

[n1ck-guo]

quantizer

Replaces the original compressor's quantize method and is responsible for the specific quantization process
Subclasses (coarse to fine granularity):

• mode (RTN, Tune): Different quantizer processes and quantize function logic.
• model_type (llm, vlm, diffusion): Different calibration methods, data processing, etc.
• data_type (gguf, mxfp8, waquanizer): Requires additional algorithms (imatrix for gguf), special processes (register_act_max_hook for WA, fused_layer_global_scale for nvfp)