Reinforcement Learning With Foundation Priors