Safe Nonlinear Control Under Control Constraints Via Reachability, Optimal Control And Reinforcement Learning