Excessive Task Diversity Inhibits Continual Reinforcement Learning