Text-guided Sparse Voxel Pruning for Efficient 3D Visual Grounding