t27b steward: tri t27b delta decides in t27 (first port slice, Refs #6198) #6202

Problem

tri t27b delta and the honest percentage in tri t27b status are decided by a hand-written Python if chain in scripts/tri_loop/t27b.py. The owner's rule of 2026-10-05 (AGENTS.md, #6199/#6200) says t27b and the tools that measure it are written in t27.

Change

• specs/tri/t27b/steward.t27: delta_code, delta_is_red, pct_tenths as integer-coded pure functions with 14 tests and 1 invariant. Passes under t27c test-report and under t27b itself (t27b test --check, 19 runtime asserts, not vacuous).
• gen/c/tri/t27b/steward.c generated by t27c gen-c (L2: never hand-edited).
• scripts/tri_loop/t27b_rules.py builds that C with the system cc into a cache and calls it via ctypes; only the JSON-string -> code mapping stays in Python (plumbing). An unknown verdict is an error.
• t27b.py delta and the percentage call it; no decision of its own left there.
• CI: scripts/ci/test_the_t27b_steward_decides_in_t27.py in loop-tools-gate, with a mutation control.
• The spec added to docs/reports/t27b_expectations.json as a pass.

Acceptance

• the loop-tools-gate step passes, mutation control red-capable
• tri t27b delta output unchanged for the same lab runs

Boundary

Only delta and the percentage. Ratchet, doctor and status reading stay in Python as recorded debt under #6198.

Refs #6198