A Preliminary Study of Preoperative Prediction of Ki-67 Expression in Gastrointestinal Stromal Tumors Using a Machine Learning Interpretable Model Integrating Enhanced CT Radiomics and Semantic Features