Continuous Vision-Language-Action Co-Learning with Semantic-Physical Alignment for Behavioral Cloning