Model-based versus Model-free Deep Reinforcement Learning for Autonomous Racing Cars