AlignCAT: Visual-Linguistic Alignment of Category and Attribute for Weakly Supervised Visual Grounding