Fix maybe_update_config signature to match updated vllm API

[Copilot]

vllm added a new hf_config parameter to QuantizationConfig.maybe_update_config, breaking the ParoQuantConfig override which only accepted (model_name, revision).

Changes

• Added hf_config: PretrainedConfig | None = None between model_name and revision in ParoQuantConfig.maybe_update_config to match the new base class signature
• Added from transformers import PretrainedConfig import

# Before
def maybe_update_config(self, model_name: str, revision: str | None = None):

# After
def maybe_update_config(self, model_name: str, hf_config: PretrainedConfig | None = None, revision: str | None = None):

The hf_config argument is accepted but unused — layer skipping continues to be auto-detected from safetensors metadata.