Look Around and Refer: 2D Synthetic Semantics Knowledge Distillation for 3D Visual Grounding