T3D: Towards 3D Medical Image Understanding through Vision-Language Pre-training