Think Too Fast Nor Too Slow: The Computational Trade-off Between Planning And Reinforcement Learning