A Convergent Reinforcement Learning Algorithm in the Continuous Case: The Finite-Element Reinforcement Learning