Splits! A Flexible Dataset and Evaluation Framework for Sociocultural Linguistic Investigation