docs: surface the v1.7→v1.12 artifact-evaluation arc in the README

[mataeil]

What

The README was last touched at v1.6.1 and never reflected the project's biggest evolution: the shift from measuring process (did a cycle advance?) to measuring whether the artifact is good.

• New section — "Measuring the artifact, not just the loop" (after "Measuring the loop"): tells the honest, probe-driven arc — Goodhart collapse → independent rubric critic → leap cycles, capture fidelity, Ambition, research-grounding, honest ceilings. Arc framing: lying A → honest D → earned A → honest F+ vs real games → honest ceiling.
• Production Validation: one clearly-labelled paragraph noting the f1 build-quality dogfood probe — explicitly not a third production deployment (it's a private-repo lab test).
• .gitignore: ignore the f1 dogfood evidence screenshots (kept on disk; never framework source) so git status stays clean.

Why

This is the most compelling part of the project's recent history (v1.7→v1.12, all merged + now tagged) and it was invisible to anyone reading the README.

Verification

• Claims adversarially reviewed in 3 lenses before shipping — factual accuracy (every score/version checked vs CHANGELOG), honesty/anti-overclaim (no rubric-score=objective-"good" conflation; probe ≠ production), and README voice/integration.
• python3 tests/verify.py → 64/64 passed, 0 failed.
• Docs-only; no engine/spec/script changes.

🤖 Generated with Claude Code