Group Masked Model Learning for General Audio Representation