A Joint Modeling of Vision-Language-Action for Target-oriented Grasping in Clutter