t27b doctor decides in steward.t27, and sees a stuck checkout #6230

Refs #6198, #6063.

Problem

tri t27b doctor still decides in Python, and three of its decisions were wrong on the live lab at 2026-10-04T18:20Z:

1. LAB-STALE reads finished only. A run that failed at checkout has no finished, so a stuck lab is never stale.
2. A failed checkout shows up as generic LAB-TESTS-RED ("checkout"), with no hint that only a redeploy (the owner's decision) or fix(t27b-lab): a clone that cannot check out is cloned again #6220's re-clone recovers it. A healed run (steps.checkout.clone == "recloned", fix(t27b-lab): a clone that cannot check out is cloned again #6220) is not reported at all (skill rule Q18).
3. Claim, railway-major, ledger-quiet and stale thresholds are Python branches.

Change

The decisions move into specs/tri/t27b/steward.t27: lab_stale, claim_code, railway_old, ledger_quiet, checkout_code, with tests. t27b.py passes them numbers and words the result. New codes: LAB-CHECKOUT (red, owner redeploy) and LAB-RECLONED (healed, report it). Stale age falls back to updated.

Acceptance

• Both t27c test-report and t27b test --check pass every test in the spec.
• The fixture test of doctor (feat(tri): tri t27b status/doctor -- the t27b steward reads its state before it acts #6114) stays green, with new cases for the three defects, and a mutation control.
• Live doctor on the current lab prints LAB-CHECKOUT instead of LAB-TESTS-RED checkout.

Boundary

scripts/tri_loop/t27b.py, scripts/tri_loop/t27b_rules.py, specs/tri/t27b/steward.t27, gen/c/tri/t27b/steward.c, scripts/ci tests, docs/now.