Collaborative Computing in Vehicular Networks: A Deep Reinforcement Learning Approach