Reinforcement learning with spiking coagents