Text this: Benchmarking large language models for biomedical natural language processing applications and recommendations