Self-Activating Neural Ensembles for Continual Reinforcement Learning