[DNM] test SGLang + LMCache server through srt-slurm

[cquil11]

Test PR, do not merge.

Runs one AgentX point through srt-slurm with SGLang's LMCache unified radix cache (enable-lmcache) against the lmcache-server service from SemiAnalysisAI/srt-slurm#32, applied as 507-lmcache-server-atom-sglang.patch.

• Qwen3.5-397B-A17B FP8, B200 TP8, conc 76, KV offload to DRAM (the HiCache c76 point with LMCache in its place).
• lmsysorg/sglang:nightly-dev-cu13-20260928-81f27fb3, LMCache 0.5.6.dev112 (nightly; the first build with the SGLang unified connector).
• No NEXTN: LMCacheUnifiedRadixCache does not support speculative decoding yet.

Result

Path	Hardware	Run	Result
lmcache-server + SGLang enable-lmcache	B200 TP8, conc 76	36490342974	Passed: 1030/1030 requests, cache hit 87% (39% GPU, the rest from LMCache)

[github-actions]

Thanks for the contribution!

• Review: If this PR changes files owned by someone other than a repository admin or @SemiAnalysisAI/core, ask one eligible CODEOWNER to complete the latest PR_REVIEW_CHECKLIST.md before contacting a core maintainer on Slack. Follow the template exactly, including As a PR reviewer and CODEOWNER, I have reviewed this and have, so sign-off verification triggers.
• PR verification: Sweeps only run on labeled PRs. Add full-sweep-fail-fast (strongly recommended); use full-sweep-enabled only when matrix jobs should continue after a failure.
• After merging: PR authors must ensure all GitHub Actions jobs pass. Transient failures often pass on rerun; see how to rerun failed jobs.

中文

感谢你的贡献！

• **审阅：**如果 PR 修改的文件归属于仓库管理员及 @SemiAnalysisAI/core 之外的 CODEOWNER，请先联系一位有资格的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，再通过 Slack 联系核心维护者。必须严格遵循模板，并保留 As a PR reviewer and CODEOWNER, I have reviewed this and have，才能触发签核验证。
• **PR 验证：**扫描仅在带有标签的 PR 上运行。强烈建议添加 full-sweep-fail-fast；仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。
• **合并后：**PR 作者必须确保所有 GitHub Actions 任务通过。临时性失败通常可以通过重新运行恢复；参见重新运行失败任务的说明。

[cquil11]

Closing: test-only PR. SGLang + LMCache server path validated on B200 (run 36490342974); result recorded in NVIDIA/srt-slurm#528.