結果 : replicable reinforcement learning with linear function approximation