Reactive Exploration to Cope with Non-Stationarity in Lifelong Reinforcement Learning