Added image-to-image task for ORT Pipeline

[h3110Fr13nd]

What does this PR do?

Fixes #2030
Image To Image task consist lot more than Diffusion and there are transformer based model such as Swin.
This PR adds the support for pielining image-to-image task with ORTModelForImage2Image.
e.g.

from transformers import AutoImageProcessor
from optimum.pipelines import pipeline
from optimum.onnxruntime import ORTModelForImageToImage
from PIL import Image

model = ORTModelForImageToImage.from_pretrained("swin2sr-2x-onnx", use_io_binding=True)
processor = AutoImageProcessor.from_pretrained("caidas/swin2SR-classical-sr-x2-64")
onnx = pipeline("image-to-image", model=model, feature_extractor=processor, device="cuda")
image_path = "image.png"
image = Image.open(image_path)
output = onnx(image)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@fxmarty, @echarlaix, @JingyaHuang, @michaelbenayoun

[IlyasMoutawwakil]

awesome ! don't hesitate to ping me for review when it's ready !

[h3110Fr13nd]

awesome ! don't hesitate to ping me for review when it's ready !

Sure. Writing some tests. Will update you soon.

[h3110Fr13nd]

Suggest Changes if Any.
@IlyasMoutawwakil, @fxmarty, @JingyaHuang, @michaelbenayoun

[IlyasMoutawwakil]

Looks good, I left a few comments and suggestions, mostly to follow the same testing pipeline we have for other ORTModels:

• base the TestCase on a simpler task (I think u based on generative tasks)
• use setup to export/use an exported model
• use tiny models for faster testing

[IlyasMoutawwakil]

LGTM !
did you test io binding locally ?

[h3110Fr13nd]

LGTM ! did you test io binding locally ?

Yes.