[common] Generalized MXFP8 fused kernels w.r.t. input tensor dimensions

[Oleg-Goncharov]

Description

Lifted restrictions on input tensor dimensions in MXFP8 fused kernels, allowing arbitrary number of rows (or the product of all tensor dimensions except the last of high-dimensional tensors).
The number of columns (or the dimensionality of the last dimension of high-dimensional tensors) must satisfy the memory alignment requirement for Tensor Memory Accelerator (TMA) to be a multiple of 16 bytes.

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

• Removed the isFullTile requirement in the mxfp8_quantize function

• Added 16-byte alignment check when creating a TMA descriptor

• MXFP8 Test Suite: Added the alignment/padding requirement for dimensions of the tensors with scaling factors to be multiples of [128,4] for row-wise and [4,128] for column-wise scaling.

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]

Mostly LGTM. It looks like this is needed before we can merge #1435.

[timmoon10]

Pipeline 23264050