Text this: Diffusion Models for Open-Vocabulary Segmentation