feat: Add secondary Nemotron perf/diagnostic patches (Liger, TP CE, FLOPs, benchmark)

[jasont314]

Adds auxiliary optimization and reporting updates separated from the core PP/EP/SQuAD correctness path.

What does this PR do ?

Introduces secondary Nemotron performance/diagnostic improvements (Liger integration path, TP CE handling, FLOPs accounting updates, and benchmark instrumentation) without changing the core PP/EP/SQuAD integration PR scope.

Changelog

• _transformers/kernel_patches.py
  • Add NemotronH-specific Liger patch path with optional RMSNorm and CE patch toggles.
• components/loss/masked_ce.py
  • Improve TP-aware masked CE behavior used in Nemotron runs.
• components/utils/flops_utils.py
  • Update FLOPs computation/reporting for better diagnostic fidelity.
• recipes/llm/benchmark.py
  • Add benchmark-side instrumentation/reporting improvements for run analysis.
• components/distributed/init_utils.py
  • Minor distributed init/reporting update used by benchmark diagnostics.
• NEMOTRON_SECONDARY_PATCH_NOTES.md
  • Add implementation notes and usage context for this secondary patch set.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• Related to # (issue)
• Validation run:
  • python -m pytest -q tests/unit_tests/_transformers/test_auto_model.py tests/unit_tests/loss/test_masked_ce.py tests/unit_tests/utils/test_flops_utils.py tests/unit_tests/recipes/llm/test_benchmark.py
  • Result: 83 passed, 7 warnings

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[akoumpa]

/ok to test 032a69a