Enhancing Interpretability of Vertebrae Fracture Grading using Human-interpretable Prototypes