[Pytorch] Fix swizzle scale output shape in variable shape case

[YangFei1990]

Description

The swizzle scale output shape computation does not respect the first/last dim settings, which still always assume even split. This PR adds support for the variable shape case.

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

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

[greptile-apps]

Greptile Summary

The PR corrects grouped MXFP8 swizzle output allocation for variable-shaped tensors by preserving the existing scale-buffer shape.

• Distinguishes variable-shaped grouped tensors from uniform-shaped tensors during scale allocation.
• Adds regression coverage for imbalanced groups, empty groups, rowwise and columnwise scales, and requantization paths.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
transformer_engine/pytorch/csrc/extensions/swizzle.cpp	Preserves scale-buffer capacity for variable-shaped grouped swizzles while retaining padded shape computation for uniform groups.
tests/pytorch/mxfp8/test_mxfp8_group_quantize_graph_safe.py	Adds focused regression cases that validate shape preservation and per-group scale contents across rowwise, columnwise, empty-group, and requantization scenarios.

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into fix_swizzle_out..." | Re-trigger Greptile}

[YangFei1990]

/te-ci L1 pytorch

[vthumbe1503]

LGTM

[YangFei1990]

/te-ci L1 pytorch