Clinician-in-the-Loop Decision Making: Reinforcement Learning with Near-Optimal Set-Valued Policies