[CI/Build][ROCm] Add CUDA/ROCm parity manifest and evidence report

[andyluo7]

Purpose

Track all 61 rows in #5731 with an executable CUDA/ROCm parity ledger instead of treating the issue bullets as a homogeneous test count.

This PR adds:

• a 61-row manifest with current CUDA source, proposed ROCm queue, owner, status, PR, exact-head build, artifact, and resolution fields;
• a JSON schema and report generator;
• a final-mode gate that rejects missing, partial, and unknown rows;
• contract tests for row completeness and evidence requirements; and
• an explicit exclusion from Buildkite pipeline-schema validation because the new YAML is metadata, not a pipeline.

The initial snapshot at upstream 78934753ff7e092b24ada0e34062cf3d121efd25 is intentionally conservative: 37 missing, 3 partial, and 21 unknown. It makes no new coverage claims. The ledger now links the first implementation set: #7604, #7605, and #7606.

Test Plan

python3 -m pytest --confcutdir=tests/buildkite -o 'addopts=' -q \
  tests/buildkite/test_rocm_parity_manifest.py
pre-commit run --files \
  .buildkite/ci/rocm_cuda_parity.yaml \
  .buildkite/ci/rocm_cuda_parity.schema.json \
  .buildkite/scripts/generate_rocm_parity_report.py \
  tests/buildkite/test_rocm_parity_manifest.py \
  tools/pre_commit/check_buildkite.py
python3 .buildkite/scripts/generate_rocm_parity_report.py --format markdown

vLLM Version: N/A; metadata and CI tooling only.

vLLM-Omni Commit: 1561e586afab1e3e6dd022c38ad9fefa6bdfb6ff

Test Result

• 5 passed
• All repository pre-commit hooks passed for the changed files.
• The generated report contains exactly 61 rows and the expected 37/3/21 missing/partial/unknown snapshot.
• git diff --check passed.

AI assistance: Used Codex to implement the manifest/report tooling, add contract tests, run the repository pre-check, and draft this PR description. I reviewed the changes and validation results.

[vllm-omni-review-bot]

This PR was classified as CI work.

CI owner: @yenuo26 @congw729 @NickCao

Routing: @yenuo26 via semantic router, CI owner, CODEOWNERS; @congw729 via CODEOWNERS; @NickCao via CODEOWNERS

@andyluo7, please review your own changes and leave a short self-review comment describing what you checked. PRs without author self-review may not be assigned a reviewer.

Please take a look when you have a chance. If you would like an automated review, mention @vllm-omni-review-bot in a comment.

[andyluo7]

Author self-review completed at exact head 1561e586afab1e3e6dd022c38ad9fefa6bdfb6ff.

I checked that:

• the manifest accounts for all 61 issue rows exactly once and keeps the initial statuses conservative;
• a row cannot be marked covered without an exact head, a PR, two passing non-soft-failed builds, and retained artifacts;
• other terminal classifications require an explicit resolution;
• the parity YAML is excluded only from the Buildkite pipeline parser because it is metadata, not an executable pipeline; and
• this PR does not change any CUDA, ROCm, or runtime execution path.

I reran the focused manifest/report tests: 5 passed. The report generator completed and git diff --check passed.

[andyluo7]

Closing this tracking-only PR after revisiting ownership and scope. The parity ledger and report are useful for AMD program management, but this change does not select tests, schedule jobs, gate pull requests, or otherwise affect vLLM-Omni CI execution. We have retained the tooling in the AMD planning repository and will publish public-safe status, implementation PR links, exact-head Buildkite evidence, and approved exceptions through issue #5731. Upstream PRs will focus on executable ROCm CI coverage. If maintainers later want repository-native machine enforcement, we can propose a smaller CI-integrated guard derived from the executable pipeline definitions.