Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation