Efficient Adaptation of Large Vision Transformer via Adapter Re-Composing