An Examination of Offline-Trained Encoders in Vision-Based Deep Reinforcement Learning for Autonomous Driving