Context-Aware Full Body Anonymization using Text-to-Image Diffusion Models