Text this: Assessing the performance of large language models in literature screening for pharmacovigilance: a comparative study