v0.7.2: fix all-black output on fp16 GPUs

Fixes

• Fix all-black output on fp16 GPUs (#29). The stock SDXL VAE overflows to NaN in fp16, so the plain img2img path could decode to a completely black image on a CUDA / XPU fp16 backend. The pipeline now loads the fp16-fixed SDXL VAE (madebyollin/sdxl-vae-fp16-fix) for the default SDXL checkpoint when running fp16. CPU and Apple MPS run fp32 and were never affected.