onnxruntime

Refine auto_pad based pad computation in ConvTranspose

#6305

Merged