Nonparametric Additive Value Functions: Interpretable Reinforcement Learning with an Application to Surgical Recovery