CLIP-Count: Towards Text-Guided Zero-Shot Object Counting