Port MI355X DeepSeek-R1 FP8 disaggregation to srt-slurm

[cquil11]

Summary

• replace the legacy MI355X DeepSeek-R1 FP8 disaggregated launcher with one native srt-slurm override recipe
• preserve all six STP/MTP topology families and every original 8k/1k concurrency point
• move to the current MI355X SGLang v0.5.17 runtime, SGLang Router, and MoRI KV transfer
• reuse the complete 688.6 GB legacy model cache through a zero-copy canonical Hugging Face cache link
• remove the retired model-specific wrapper and keep benchmark_serving.py unchanged via the custom benchmark interface

Topologies

• STP and MTP1: 2P DEP8/DPA + 1D DEP8/DPA at c1024/c2048
• STP: 1P TP8 + 2D TP8 at c4-c256
• MTP2: 1P TP8 + 2D TP8 at c2-c256
• STP and MTP2: 1P TP4 + 1D TP8 at c1-c64

Validation

• all six resolved recipe selectors pass srt-slurm dry-run at c609754b5622f96d5c12a93149e245308d4f1e9b
• resolved topology, parallelism, MTP, MoRI dispatch, and benchmark environment assertions pass
• 231 matrix generation and validation tests pass
• exact six-job InferenceX matrix generation passes
• launcher bash syntax and repository diff checks pass
• cluster audit confirms 163/163 indexed model shards with zero missing files

Stack

Based directly on #2542. The shared MI355X launcher, cluster profile, and legacy cleanup live only in the base; this PR contains only the DeepSeek-R1 FP8 matrix entries, recipe, and changelog entry.

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档