Deep Correlated Prompting for Visual Recognition with Missing Modalities