[Common] Pack attention arguments as structs

[cyanguwa]

Description

Please include a brief summary of the changes, relevant motivation and context.

Fixes # (issue)

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

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

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

[timmoon10]

Instead of exposing all of the options in a struct, it may be worth considering putting them in an opaque object like NVTEQuantizationConfig:

TransformerEngine/transformer_engine/common/include/transformer_engine/transformer_engine.h

Line 395 in b38c16b

typedef void *NVTEQuantizationConfig;

This has a few advantages:

• The struct definition is fully within the core lib, so it's more robust to cases where libtransformer_engine.so and transformer_engine_torch are from different TE versions.
• Default initialization is automatic and doesn't require the non-standard init macros.
• It's easier to deprecate options. Remove the option from the internal struct, and either error out or ignore if the user attempts to set it.