MoManipVLA: Transferring Vision-language-action Models for General Mobile Manipulation