Bootstrapping Reinforcement Learning with Imitation for Vision-Based Agile Flight