Discounted Reinforcement Learning is Not an Optimization Problem