[codex] support optional heterogeneous hardware metadata

[cquil11]

Adds optional hardware metadata inside the prefill and decode worker blocks for heterogeneous disaggregated multi-node configs. The fields must be specified together or both omitted; omission preserves the existing homogeneous behavior. disagg: true is rejected for single-node configs.

Worker metadata flows through generated matrices and workflows into both regular and agentic aggregate results as prefill_hw and decode_hw. It does not affect scheduling or launcher behavior, and no existing master config entries are modified.

Validation: 186 matrix-logic tests, 28 regular process-result tests, and 26 agentic process-result tests pass; all 226 active master entries validate successfully.

Follow-up: #2105 tracks consolidating duplicated regular/agentic metadata processing into a shared helper.

[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

[claude]

Claude finished @cquil11's task in 2m 49s —— View job

---

PR Review: heterogeneous hardware metadata

• Gather context and read the PR diff
• Review validation logic, workflow plumbing, and result processing
• Post review feedback

LGTM - no blocking issues found

Verified during review:

• All seven benchmark-multinode-tmpl.yml call sites receive the new prefill-hardware/decode-hardware inputs — the sweep-multi-node-8k1k job (.github/workflows/run-sweep.yml:405) inherits them via the *multi-node-inputs YAML anchor, so nothing was missed.
• When a config omits hardware, ${{ matrix.config.prefill.hardware }} resolves to an empty string, which matches the template's default: "" and the empty-string handling in both result processors, so homogeneous behavior is preserved end-to-end.
• The must-be-specified-together (XOR) check is enforced consistently in three places: the pydantic models (utils/matrix_logic/validation.py:129), utils/process_result.py:71, and utils/agentic/aggregation/process_agentic_result.py:72.
• Tightening disagg to Literal[False] on single-node matrix/master models is safe — YAML booleans parse to Python bool, no existing master entries change, and the author confirmed all 226 active entries validate.
• The direct _gpu_shape() unit test is safe despite not setting the other multinode env vars, since env_int defaults to 0.

Notes per review checklist: no master configs were modified (no perf-changelog entry needed), and generate_sweep_configs.py was not touched (no line count report required). Test suites could not be re-run in this sandbox (pytest execution not permitted), but the changes were verified by inspection and the PR's CI covered them.