CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure