Learning from Children: Improving Image-Caption Pretraining via Curriculum