MLREF: Efficient Module Reuse for Reward Design in Reinforcement Learning via Large Language Models