Processing speed enhances model-based over model-free reinforcement learning in the presence of high working memory functioning