feat(cli): `tirith test` — a policy test harness (skeleton)

There is no way to assert that a policy does what its author thinks; verifying the examples means
hand-running one invocation per fixture and reading exit codes. Add a fixture convention —
<policy>.fails.json, <policy>.passes.json, <policy>.skips.json next to the policy — and a
runner: tirith test examples/ prints one ✓/✗ per (policy, fixture) with expected vs actual exit.
Land the skeleton in R1 so the R1 fixes ship with fixtures; the full version (CI wiring, docs) is
the R2 issue "complete the policy test harness". Nothing else on the roadmap compounds like this:
without it, community rules cannot be reviewed.

Start at the CLI entry point for the tirith test command and inspect how policies and examples are currently invoked. Implement the R1 skeleton around the <policy>.fails.json, <policy>.passes.json, and <policy>.skips.json fixtures in examples/, printing one result per policy and fixture with expected versus actual exit codes.