Can You Turn Your Vision-Language Model from a Zero-Shot Model to Any-Shot Generalist? Meet LIxP, the Context-Aware Multimodal Framework