Phase 13 prep: ollama provider + gpt-oss:120b baseline eval (3 seeds × 6 scenarios)

Phase 9: baseline eval harness (heuristic + LLM agents) + tests