Point Cloud Completion with Pretrained Text-to-Image Diffusion Models