A Reinforcement Learning Approach to Non-prehensile Manipulation through Sliding