[Pytorch] Optimize CPU overhead for NCCL EP eager mode

[YangFei1990]

Description

Reduce NCCL EP CPU overhead by using pinned memory CPU tensor.

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

[YangFei1990]

/te-ci L1 pytorch

[greptile-apps]

Greptile Summary

The PR reduces eager-mode NCCL expert-parallel CPU overhead by storing the per-step receive-token count in pinned host memory.

• Keeps graph-mode receive-token accounting on the CUDA device.
• Synchronizes the active CUDA stream before reading the eager-mode host counter.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
transformer_engine/pytorch/ep.py	Uses a pinned host receive counter in eager mode and synchronizes before host-side output sizing; no eligible follow-up defect was established.

_{Reviews (2): Last reviewed commit: "Merge branch 'main' into ncclep_cpu_over..." | Re-trigger Greptile}