[Core] CuteDSL MoE with Nvfp4 DeepEP dispatch

[wenscarl]

Purpose

Dispatch with nvfp4 DeepEP low latency mode.
deps on deepseek-ai/DeepEP#341 and #25990

Test Plan

VLLM_USE_FLASHINFER_MOE_FP4=1
VLLM_USE_STANDALONE_COMPILE=0
VLLM_FLASHINFER_MOE_BACKEND="cutedsl"
VLLM_WORKER_MULTIPROC_METHOD=spawn
VLLM_ALL2ALL_BACKEND="deepep_low_latency"
lm_eval --model vllm --model_args pretrained=nvidia/DeepSeek-R1-FP4,data_parallel_size=4,enable_expert_parallel=True,tensor_parallel_size=1,enforce_eager=True,max_model_len=2048 --trust_remote_code --tasks gsm8k --num_fewshot 5 --batch_size auto

Test Result

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
gsm8k	3	flexible-extract	5	exact_match	↑	0.9462	±	0.0062
		strict-match	5	exact_match	↑	0.9462	±	0.0062

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.