Watch, Try, Learn: Meta-Learning from Demonstrations and Reward