BUSTR: Breast Ultrasound Text Reporting with a Descriptor-Aware Vision-Language Model