Support W8A8_dynamic on Step3 Model

[yifeilin1015]

###What this PR does / why we need it?
Support w8a8_dynamic for Step3 model

###Does this PR introduce any user-facing change?
No user-facing change

###How was this patch tested?
This patch is tested with following datasets:
MMMU: 73.11， MMSTAR: 70.68, AIME25: 80.00， GPQA-diamond: 72.73

###Modify points:

• add step3_text.py file with new load_weights method
• load_weights method only change the way of loading MoE experts compared to the original method

vLLM version: v0.10.2

• vLLM version: v0.10.2
• vLLM main: vllm-project/vllm@f2718d2

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request adds support for the Step3 model with W8A8 dynamic quantization. The changes include registering the new model and implementing a custom load_weights method to handle the specific weight structure of MoE experts in this model. My main feedback is to avoid hardcoding the number of experts to improve the model's reusability.

[gemini-code-assist]

The number of experts is hardcoded to 48, which limits this implementation to a specific model configuration. To support Step3Text models with a varying number of experts, experts_ and packed_modules_mapping should be initialized as instance attributes within __init__. This allows dynamically setting the number of experts from the model configuration (self.model.config.moe_num_experts).

    def __init__(
        self,
        *,
        vllm_config: VllmConfig,
        prefix: str = "",
    ):
        super().__init__(vllm_config=vllm_config, prefix="model")
        num_experts = self.model.config.moe_num_experts
        self.experts_ = [f"experts.{i}.{proj}" for i in range(num_experts) for proj in ("down_proj", "gate_proj", "up_proj")]
        self.packed_modules_mapping = {
            "qkv_proj": [
                "q_proj",
                "k_proj",
                "v_proj",
            ],
            "gate_up_proj": [
                "gate_proj",
                "up_proj",
            ],
            "experts": self.experts_,
        }

[realliujiaxu]

this should be moved this to quant_config.py, after https://github.com/vllm-project/vllm-ascend/pull/3021/files is merged