Supervised fine-tuning of pre-trained antibody language models improves antigen specificity prediction