ci(pitot): surface per-agent E2E status

[bigboateng]

Why

Pitot's coding-agent E2E coverage ran inside one broad Intelligence Flow job, so reviewers and public Pitot users could not see independent Claude, Cursor, Codex, Gemini, and OpenCode health.

What changed

• Add one reusable Ubuntu/macOS E2E matrix with five agent-specific workflow entrypoints.
• Emit strict JSON evidence that distinguishes real CLI execution from active hook-subprocess verification.
• Add a trusted workflow_run reporter that maintains one sticky evidence table on future Pitot PRs without executing PR code.
• Carry the five native Intelligence Flow main workflow badges into the projected public Pitot README.
• Keep all E2E execution and reporting infrastructure in Intelligence Flow; the public Pitot repository receives only the projected README.

Verification

• python3 -m unittest discover -s labs/15-pitot/tests -p 'test_*.py' — 18 passed
• python3 labs/15-pitot/scripts/build_pitot.py --repo . --check
• go test ./labs/15-pitot/pitot/...
• go test -tags windtunnel ./labs/15-pitot/pitot/windtunnel/
• bash labs/15-pitot/scripts/check_import_boundary.sh
• actionlint across the Pitot workflows
• Local E2E scripts: Claude real CLI; Cursor, Codex, Gemini, and OpenCode hook verified

Bootstrap note

GitHub only runs a workflow_run reporter after that workflow exists on the default branch. This bootstrap PR will run the five agent checks, but its sticky PR comment will not appear. The reporter becomes active for subsequent Pitot PRs after merge. README badges resolve after the first corresponding main runs complete.

Rollback

Revert this change to restore the combined E2E loop and remove the projected badges and reporter.

[bigboateng]

Pitot coding-agent E2E — manual backfill

PR #102 introduced the trusted sticky reporter, so GitHub could not run that workflow_run reporter until it existed on the default branch. The per-agent E2E checks implemented by this PR covered Ubuntu and macOS; Windows agent E2E was not part of the matrix.

Agent	Ubuntu	macOS	Windows	Result
Claude	✅ Pass	✅ Pass	⚪ Not run	✅ Passing on implemented platforms
Codex	✅ Pass	✅ Pass	⚪ Not run	✅ Passing on implemented platforms
Cursor	✅ Pass	✅ Pass	⚪ Not run	✅ Passing on implemented platforms
Gemini	✅ Pass	✅ Pass	⚪ Not run	✅ Passing on implemented platforms
OpenCode	✅ Pass	✅ Pass	⚪ Not run	✅ Passing on implemented platforms

Source commit: 7d229f9a929f

The separate Runtime (windows-latest) check passed, but it does not exercise each coding-agent integration and is therefore not reported as Windows agent E2E evidence. Future Pitot PRs can use the automatic reporter now that the workflow is present on main.