Homeostatic reinforcement learning for integrating reward collection and physiological stability | Synapse