Comparing computable structured phenotype- versus large language model-identification of opioid use disorder using electronic health record data.