Benchmarking Large Language Models for Italian Medical Text Classification: Are Generative Models the Best Choice? — SciRadar