Semantic-Clipping: Efficient Vision-Language Modeling with Semantic-Guidedd Visual Selection