Self-supervised machine learning using adult inpatient data produces effective models for pediatric clinical prediction tasks