Reinforcement Learning and Regret Bounds for Admission Control