docs(eval): add README for running evals locally

[fullsend-ai-coder]

Summary

• Add eval/README.md documenting how to run agent evals locally
• Covers prerequisites (tools, GCP credentials, GitHub token scopes), required env vars, run commands, test case structure, and known issues
• Documents both review and triage eval workflows since they share the same infrastructure

Closes #247

Test plan

• Verify eval/README.md exists and renders correctly
• Confirm all required env var names from eval.yaml configs are documented
• Confirm run command matches eval/run-functional.sh usage
• Confirm known issues section references Reintroduce review eval tests — self-review token blocker #245 and UploadFile fails when remotePath basename matches source filename under /tmp/ fullsend#5231
• Verify no secrets or credentials are present in the file

🤖 Generated with Claude Code

---

Closes #247

Post-script verification

• Branch is not main/master (agent/247-eval-readme)
• Secret scan passed (gitleaks — def8cb529ebf40b98eb0c0b67a2793b46f7648e4..HEAD)
• PR body secret scan passed (gitleaks — no-git)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox