CLIP-Forge: Towards Zero-Shot Text-to-Shape Generation