One Ref: Unified One-tower Expression Grounding and Segmentation with Mask Referring Modeling