Eval set for design modes and concepts

Imported from GitHub issue El-SaMa/tekori#16 by @El-SaMa.

---

Tasks

• scripts/eval/briefs.json: 24 briefs (8 business kinds × 3 moods) with expected checklist items; each run in safe, bold, experimental.
• Scoring: proof checks, checklist coverage (auto), template score (#13), design score, cost, time.
• Mechanics run on gpt-4.1-nano or local (LM Studio / qwen) only: wiring works end to end.
• Paid comparison is blocked until the operator approves a budget (rule: no TEKORI credit burn). Prepare the exact run command and cost estimate in the issue when ready.

Acceptance criteria

• Results table committed under scripts/eval/results/; summary posted here.

Depends on

#1-#15.

---

Plan key T16. Epic: El-SaMa/tekori#19.