Add race condition flakiness detection agentic workflow

[Copilot]

Adds a dedicated gh-aw agentic workflow to surface tests that fail intermittently due to concurrency bugs, distinct from the existing general flaky-test detection.

New workflow: race-condition-detection

• Source + lock: .github/workflows/race-condition-detection.md compiled to race-condition-detection.lock.yml.
• Trigger: weekday-only schedule cron: "0 7 * * 1-5" plus workflow_dispatch.
• Classification: flags a test as a race-condition suspect on either signal:
  • Same-commit non-determinism — the same test passes and fails across runs sharing one headSha.
  • Concurrency signature — failure type/message matches a per-language indicator (Go DATA RACE, Java ConcurrentModificationException, ThreadSanitizer, deadlocks, PoisonError, asyncio/await timing, etc.). Deterministic failures are explicitly out of scope.
• Inputs: reuses the proven artifact-download + per-language analyzer steps: from corn-flakes-detection, so the agent only reads pre-generated ./reports/*.md (the sandbox blocks python3). runs.json carries headSha to enable same-commit grouping.
• Outputs: one issue per suspect via safe-outputs (create-issue prefix [race condition] , labels race-condition,flaky-test,automated; update-issue; noop), including a PR-title remediation instruction for the Copilot coding agent.

Security posture

• Least-privilege read-only permissions, network: {}, GitHub MCP toolset for reads (no raw curl).
• All actions SHA-pinned; github/gh-aw/actions/setup pinned to f990bbb7…, matching sibling workflows and the repo's full-SHA pin rule.

Notes for reviewers

• The GH_TOKEN env warning at compile time is shared with corn-flakes-detection (read-only gh access for the agent) and is the reason strict: false is retained.
• Azure MCP (per issue guidance) is intentionally not wired in — this workflow operates solely on GitHub Actions test artifacts with no Azure infrastructure, so adding it would introduce secrets/attack surface for no functional gain.