small improvements in diffusion pipeline for openvino

[eaidova]

What does this PR do?

This PR propose small improvements for OVStableDiffusionPipeline

1. When diffusion pipeline compiled it logs message about compiling encoder 3 times, which may confuse users. Added name of model part in logs to make these logs more useful (e.g. in case if one of the models filed during compilation, we can understand which one).
2. Model with static shape does not allow to change shape during pipeline call, but user should know current height and width to understand when model requires reshaping. this PR proposes properties height and width where user can check current pipeline status
3. Fixed export models using torch 2.0.0

[eaidova]

@AlexKoff88 @helena-intel could you please take a look?

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[helena-intel]

Thanks @eaidova ! Could you also create a test for this in https://github.com/huggingface/optimum-intel/blob/main/tests/openvino/test_modeling.py ?

[eaidova]

Thanks @eaidova ! Could you also create a test for this in https://github.com/huggingface/optimum-intel/blob/main/tests/openvino/test_modeling.py ?

@helena-intel, thank you, added

[eaidova]

@AlexKoff88 @helena-intel @echarlaix

I observe failing on ci, because stable diffusion onnx export failed with torch 2.0.1 (aten::scaled_dot_product_attention is not registered for export. It will be added in next release as it is already exists on pytorch master and in nightly packages.
I added registering custom operation for that as workaround, please take a look, if it is not acceptable, I can remove it

[AlexKoff88]

@echarlaix, can you please take a look what is wrong with the documentation build?

[echarlaix]

Thanks a lot @eaidova !

@echarlaix, can you please take a look what is wrong with the documentation build?

The documentation generation was fixed in #343