Fast Lifelong Adaptive Inverse Reinforcement Learning from Demonstrations