[Feature]: Support Omni model quantization

[xin3he]

Feature Description

Current, Omni model is becoming popular, but AutoRound is based on traditional LLM structure and cannot handle it.
We need to support this structure, at least quantize Qwen3OmniMoeThinkerTextModel and Qwen3OmniMoeTalkerModel sucessfully.

Motivation and Use Case

from auto_round import AutoRound

model_name_or_path = "Qwen/Qwen3-Omni-30B-A3B-Instruct"

ar = AutoRound(
    model=model_name_or_path,
    scheme="W4A16",
    iters=50,
    lr=5e-3,
)
ar.quantize_and_save(format="auto_round", output_dir="tmp_w4a16")

Related to: #775, #862, #1186

Alternatives Considered

No response

Definition of Done

No response

Additional Context

No response

[lvliang-intel]

Addressed an issue in transformers (#43684) regarding Omni model support. A monkey patch need to be applied to the config class to resolve the initialization order issue, as per the ongoing implementation in PR #43593. I will try to raise one PR to support this model quantization.