VQC-Based Reinforcement Learning with Data Re-uploading: Performance and Trainability