Add NeuronModel classes for inference on nlp tasks

[JingyaHuang]

• Add NeuronModel base class architecture.
• Add NeuronModelForXXX subclasses for tasks: feature-extraction, fill-mask, multiple-choice, question-answering, text-classification, token-classification.
• Support padding when input shapes is not equal to the compilation static shapes.
• Add tests for modeling
• Update precompiled examples to HF hub

Upcoming

• Maybe cache for each export (so far neuron compiled models won't be cache unless it exits on the hub)
• Some models have accuracy issues that need to further investigate (c.f comments in test_modeling.py)
• All remaining tests passed for neuronx on INF2, need to test with INF1 and Trainium, might be some patches.
• Add documentation for NeuronModel(s)

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[philschmid]

Thats an awesome but big PR! Didn't went into detail on all of the code changes.
Can we add a documentation Page for similar to https://huggingface.co/docs/optimum/onnxruntime/usage_guides/models as a guide and
the classes as reference: https://huggingface.co/docs/optimum/onnxruntime/package_reference/modeling_ort

[JingyaHuang]

Thats an awesome but big PR! Didn't went into detail on all of the code changes. Can we add a documentation Page for similar to https://huggingface.co/docs/optimum/onnxruntime/usage_guides/models as a guide and the classes as reference: https://huggingface.co/docs/optimum/onnxruntime/package_reference/modeling_ort

@philschmid I will add the doc in a separate PR as this one is huge enough.

[JingyaHuang]

Issue opened for logging performance problems that we found in this PR: #90

[JingyaHuang]

The suggestions are addressed, gently ping you for a final check then I am gonna to merge it. @michaelbenayoun

[michaelbenayoun]

LGTM!
Huge work @JingyaHuang

[philschmid]

Congrats 👯‍♀️ @JingyaHuang