[Multi-Modifier] Scoped apply quantization config

[brian-dellabetta]

In order to support multi-modifier recipes (e.g. AWQ+W4A16 on self_attn layers and FP8_DYNAMIC on mlp layers), quantization config and status must be applied only to the modules scoped to the modifier, not all at once. This updates apply_quantization_config so that quantization_config and quantization_status are applied just to the target modules, not changed globally across all modules.

In order for proper target prioritization, apply_quantization_status is performed regardless of what the current status is for the model. Without these changes, test_target_prioritization will fail.

Other small changes:

• Added a test_multi_apply_quantization_config to make sure the application of multiple quantization configs in series works correctly -- shapes are correct and unused parameters are correctly removed.
• Drop override_quantization_status in favor of more general patch_attr.
• Removed infer_quantization_status which is no longer meaningful at the model level. It is also no longer needed because module's current status isn't checked.
• Added ALL_QPARAM_NAMES constant so that parameters related to quantization can be cleared from modules during init
• Removed all references to "quant_method": "sparseml" in favor of "compressed-tensors"
• Dropped usage of compress_quantized_weights and apply_quantization_status. We can remove compress_quantized_weights and references to it in examples/notebooks in a follow-up PR
• Also updated tests to get rid of warnings:

tests/test_compressors/quantized_compressors/test_fp8_quant.py::test_quant_format[channel-None-sc2-zp2]
  /home/runner/work/compressed-tensors/compressed-tensors/tests/test_compressors/quantized_compressors/test_fp8_quant.py:78: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
    "dummy.weight_scale": torch.tensor(sc, dtype=torch.float32),

Merge in conjunction with

• [Multi-modifier] Support scoped application of quantization config/status vllm-project/llm-compressor#1772

[kylesayrs]

FYI #428. Also touches some apply logic and adds more scheme merging

[kylesayrs]

I don't think a warning is necessary if the schemes overwrite. Looks good to me

[kylesayrs]

LGTM!

[rahul-tuli]

Good job! LGTM! 🚀