fix: broaden JSON answer extraction for model formatting variations

feat: ethical refusal as scored dimension, not failure

feat: generative toxigen eval with silent exit detection

feat: add quick_eval.py + 4 curated benchmark JSONL sets