Dynamic Distribution Adaptation in Transfer Learning for Cross-Project Just-in-Time Defect Prediction: A Synergistic Approach Incorporating Kernel Variance Matching, Correlation Alignment, and Categorical Boosting