[Common] Store swizzled MXFP8 rowwise scales directly in the register-resident

[kainzhong]

TODO: let me review it

The GEMM-swizzled path of quantize_contiguous_kernel gathered each 16-byte scale run through warp shuffles and shared memory, then flushed it after a CTA barrier. Store each scale byte directly at its swizzled index instead, as the packed path does. Outputs are bit-identical; swizzled bf16 rowwise quantization is ~2.4% faster on average (up to ~5.5%) on GB200.

Description

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

Benchmark: https://claude.ai/artifact/VpCXcAtAvbCe7gfxWBEwLV?sk=XE0XRwAptECOploI4sXZeA

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