fix(stargate-bench): measure validated stream output

[barrygreengus]

Why

The benchmark driver treated an empty 200 response as success and the scorer credited requested output tokens as generated output. This could report output throughput without completed inference output.

What changed

• Validate bounded SSE events and require the stream's completion marker.
• Read reported completion-token usage and request usage reporting from the backend.
• Keep requested workload tokens separate from observed output counts.
• Mark throughput and output shares unavailable when successful responses lack current usage, including legacy result files without the new field.
• Add the benchmark suite to Bazel with declared scenario runfiles.

Customer Release Notes

Stargate benchmarks report observed output usage and reject empty or truncated success responses. Missing usage is reported as unavailable instead of estimated from the request.

Plan Summary

Not applicable.

Usage

Request results include observed_output_tokens. Existing output_tokens fields retain the requested workload count. Older result files remain readable; output metrics are unavailable when observed usage is absent.

Testing

The rebased Cargo workspace passed all 1,611 tests, with 3 existing ignored performance tests. Workspace Clippy passed for all targets with warnings denied. All 161 benchmark tests also passed through Bazel with declared scenario runfiles. Coverage includes empty and truncated responses, fragmented UTF-8, role-only chunks, observed usage, stale counters, missing usage, and legacy result files. No GPU or live-cluster performance campaign was run.

Notes

Final layer of the Stargate maintenance stack. Earlier layers fix protocol handling, proxy and relay lifetime, pylon request ownership, and registration recovery. The streaming deadline finding was already fixed by #1452 and verified by the workspace tests.

Issues

Closes #1817

References

None.

Related Pull Requests

Depends on #1823. Stack, bottom to top: #1818, #1820, #1822, #1823, #1826.

Dependencies

No added or upgraded dependencies. No license or NOTICE changes.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🛡️ CodeQL Analysis

🚨 Found 5 issue(s)

Severity Breakdown:

• 🔴 Errors: 0
• 🟡 Warnings: 0
• 🔵 Notes: 0

📋 Top Issues

🔗 View full details in Security tab

_{🕐 Last updated: 2026-09-11 22:27:38 UTC | Commit: 54b3e6e}