test(runner): establish compatibility baseline

[BjRo]

Why

Closes #211. Darrow needs executable behavior-parity evidence before generic evaluation infrastructure moves to Sevro under ADR-0010.

What changed

Adds a configurable command-level suite that runs the current eval runner in an isolated project with a credential-free synthetic adapter. The baseline covers case selection, project and result roots, success and failure categories, result and diagnostic artifacts, cancellation, and retained trial evidence without asserting terminal wording or runner module structure. It also documents bun run test:eval-runner-compatibility and the JSON-argv override for a future Sevro launcher.

Verification

• bun run test:eval-runner-compatibility — 5 passed
• bun test evals/runner/run-cli.test.ts evals/runner/run-selection.test.ts evals/runner/run-artifact.test.ts evals/runner/run-recovery.test.ts — 43 passed
• bun test evals/runner/suite.test.ts — 8 passed on isolated rerun
• bun run typecheck — passed
• bun run lint:ts — passed
• bun run lint — passed
• bun run lint:shell — passed
• bun run check:docs — passed
• bun run check:decisions — passed
• bun run check:python — passed
• bun test — 635 passed; one unchanged discovery-eval test failed reproducibly, and one suite test timed out under the full parallel load but passed on the isolated rerun above

Review notes

The alternate-runner seam is a JSON argv array with fixture-path placeholders; the configured launcher owns synthetic-adapter wiring for the four documented scenarios. This slice does not move runner source, change Sevro, publish a package, or switch Darrow evaluation commands. The full-suite discovery failure is in unchanged eval data and also fails when run alone.

Checklist

• I have read and followed CONTRIBUTING.md, including the contribution
  licensing terms.
• I added or updated the applicable invariant before implementation, or
  this change does not affect a capability invariant.
• I added or updated colocated evals, or this change does not affect skill
  behavior.
• I confirmed that each changed plugin remains self-contained, or this
  change does not affect plugin content.
• I ran bun run check:python, or this change does not affect registered
  Python packages or their repository quality infrastructure.