Stable and Safe Reinforcement Learning via a Barrier-Lyapunov Actor-Critic Approach