Zero-Shot Policy Transfer in Autonomous Racing: Reinforcement Learning vs Imitation Learning