Safe Reinforcement Learning on FormulaOne L2 (held-out seeds 10000–10019)