Reinforcement Learning with Linear Logic Guidance