Disable graph check while tracing

[mvafin]

What does this PR do?

This is alternative to #1064. It will only work with openvino>=2025.0 (openvinotoolkit/openvino#28328)
By disabling checking graph after trace we reduce memory usage in diffusers.

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?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[IlyasMoutawwakil]

I would prefer if we used something like TorchScriptPythonWrapper or TorchScriptPythonModel, the decoder part doesn't make sense especially since we are using this wrapper on top of diffusers text_encoder/unet/transformer etc.
Will this wrapper ever change and do decoder-specific stuff ? if npt why does it have Decoder in its name.

[mvafin]

This is a "decoder" in a sense that it is used to decode the pytorch model. It is used in openvino as an abstraction to get the graph from the model. It is an internal class, but it can be used to provide such custom parameters for tracing.

[eaidova]

this naming does not have relation to decoder models, Decoder is internal component of openvino responsible for decoding of original framework model during conversion. All that time we use it internally inside convert_model, now it is created explicitly for resolving advanced conversion options

[mvafin]

Well, we can rename it on import. Will TorchScriptPythonWrapper be a better name?

[IlyasMoutawwakil]

oh okay, confusing name but okay for me if it's part of the openvino naming conventions.

[AlexKoff88]

@IlyasMoutawwakil, can we merge this then?

[mvafin]

Per @eaidova ask I moved decoder import inside function, can be useful in case openvino was custom built without pytorch support

[IlyasMoutawwakil]

LGTM, can merge once tests pass.