Add opt-in reduced precision output for cuDNN MXFP8 norm

[sraman-rgb]

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

[greptile-apps]

Opt-in path lacks coverage

The existing MXFP8 normalization tests leave NVTE_CUDNN_MXFP8_NORM_OUTPUT_IN_INPUT_DTYPE disabled, so CI does not exercise graph construction or output correctness when this branch changes the virtual normalization output from FP32 to FP16 or BF16.

Knowledge Base Used:Common core (transformer_engine/common)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

Greptile Summary

Adds an opt-in reduced-precision intermediate output for cuDNN MXFP8 normalization.

• Gates the behavior on cuDNN 9.25+, one-dimensional block scaling, and a new environment variable.
• Uses the input FP16/BF16 datatype for the virtual normalization output while retaining the requested final quantized output type.
• Documents the new environment variable and its fallback behavior.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
transformer_engine/common/normalization/common.cpp	Adds version-, scaling-mode-, and environment-gated selection of the input datatype for the cuDNN MXFP8 normalization intermediate output.
transformer_engine/common/normalization/common.h	Declares the helper that reads the new MXFP8 normalization environment setting.
docs/envvars.rst	Documents the new opt-in setting, supported cuDNN version, datatype behavior, and default fallback.

_{Reviews (4): Last reviewed commit: "Scope cuDNN MXFP8 dtype override to FP8 ..." | Re-trigger Greptile}

[vthumbe1503]

LGTM