[FEAT] [ROCm]: Add Qwen/Qwen3-235B-A22B-FP8 TP4 triton fused moe config

[tjtanaa]

Add Qwen/Qwen3-235B-A22B-FP8 TP4 triton fused moe config

VLLM_ROCM_USE_AITER=0 VLLM_USE_TRITON_FLASH_ATTN=0 python3 benchmarks/benchmark_throughput.py --input-len $_in_len --output-len $_out_len --trust-remote-code --num-prompts 200 --model Qwen/Qwen3-235B-A22B-FP8 --max-model-len 32768 --gpu_memory_utilization 0.95 --tensor-parallel-size 4 --max_seq_len_to_capture 32768 --quantization fp8 --kv-cache-dtype fp8

Input Tokens	Output Tokens	Metric	Before Tuning	Tuned Config	% Gain
5000	1000	Requests/s	0.96	1.03	+7.29%
		Total Tokens/s	6014.29	6503.40	+8.13%
		Output Tokens/s	956.43	1032.81	+7.98%
1000	2000	Requests/s	1.01	1.18	+16.83%
		Total Tokens/s	3094.05	3609.24	+16.65%
		Output Tokens/s	2020.06	2358.50	+16.76%
10000	500	Requests/s	0.69	0.71	+2.90%
		Total Tokens/s	7674.34	7892.46	+2.84%
		Output Tokens/s	346.87	356.83	+2.87%

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀