[Pytorch] Dynamo ONNX export support

[pggPL]

Description

This PR adds torch onnx dynamo-based export support. This draft is published to discuss the direction of the progress, it is not functional yet.

Current idea:

• define onnx_forward in modules, which is much simpler version of forward, which will be run in onnx_export mode. Idea: it was hard to make dynamo work with original layers and inference flow is much much simpler, so this is not painful,
• I implement many parts in onnx_export using torch modules, which work with onnx export - like layer norm,
• I implement onnx gemm and onnx quantization functions in Python and I register them in torch.ops,
• I put mapping torch.ops call <-> symbolic function inte_translation_table, which need to be passed to torch.onnx.export(dynamo=True, custom_translation_table=te_translation_table).
• torch.onnx.export(dynamo=True) seem to be working for some modules, torch.onnx.dynamo_export will not work, because it has unsupported features like generators, which are essential part of the TE layers,

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[borisfom]

I don’t like onnx_forward idea, we used similar design for Nemo and it’s not something I am proud of …
Now if we do indeed have a need to use simplified versions for forward for ONNX export, I would suggest using monkeypatch techniques to temporarily modify exported objects just for ONNX export.
Putting is_in_onnx_export_mode conditional blocks permanently in the code may cause all sort of issues with other types of export clients may need : jit.script(), torch.compile(), torch.export()