Text this: Addressing selection biases within electronic health record data for estimation of diabetes prevalence among New York City young adults: a cross-sectional study