onnxruntime

UNet fusion and fp16 conversion for stable diffusion (#14248)

Add script to fuse nodes to optimized operators in stable diffusion 1.5
models, and a script to convert fp32 models to fp16 models. Tested with
stable diffusion 1.5.

Note that the optimized model needs onnxruntime-gpu v1.14 (release candidate
will be available soon).

Note: We will update the script to work with latest diffusers and stable
diffusion v2 and v2.1 models.