agent-team: reviewer-agent should run the test command and fail-closed on skip, not just parse the PR body

[verkyyi]

Summary

Reviewer prompt today says "Re-run the tests yourself if feasible to confirm" and grades ## Test status from the PR body. In the verkyyi/agentfolio#105 review (run 24682194207), tests weren't re-run — the reviewer trusted the implementer's self-report. For single-contributor agent-team runs this works (the implementer is honest), but the system is only as trustworthy as its weakest loop. Trusting self-report undermines the whole point of a review gate.

Proposal

• Reviewer agent detects the test command by reading package.json / Makefile / pyproject.toml / go.mod / existing CI files, in that order. If found, it must run it against the PR branch checkout.
• If the repo has no detectable test command: mark ## Tests as ⚠ no test command detected — skipped by design and proceed.
• If detection succeeds but the command errors (real failure, not infra): verdict is kickback regardless of other checks.
• If detection succeeds but command is skipped without a detectable reason: verdict is block.

Acceptance

• reviewer-agent.md prompt gates approve on a successful test run.
• Reviewer ## Review comment includes the exact command and a ✅/❌ outcome.
• catalog/agent-team/README.md documents the test-command detection order and how to override via .agent-team.yml (see sibling issue on per-repo config).