[Pytorch] [NCCL EP] Allow zero tokens for an EP rank in eager mode

[YangFei1990]

Description

For eager mode, there could be case where certain EP ranks do not receive any token, based on the routing strategy. This PR relax the constraint to allow such behavior.

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

[YangFei1990]

/te-ci L1

[greptile-apps]

Greptile Summary

Updates expert-parallel eager execution to accept empty tensors with null data pointers.

• Relaxes the NCCL EP tensor descriptor validation only for zero-element tensors.
• Adds a distributed eager-mode test covering zero-token receive ranks through dispatch, combine, and backward.
• Updates the NCCL extensions submodule revision.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
transformer_engine/common/ep/ep_backend.cpp	Allows null tensor data only when the corresponding NVTETensor contains zero elements.
tests/pytorch/distributed/run_ep.py	Adds eager-mode distributed coverage for zero-token receive ranks across forward and backward execution.
3rdparty/nccl-extensions	Advances the NCCL extensions submodule revision supporting the updated EP behavior.

Sequence Diagram

sequenceDiagram
  participant Router
  participant Dispatch as EP Dispatch
  participant Rank as Zero-receive EP Rank
  participant Combine as EP Combine
  Router->>Dispatch: Route all tokens to another rank
  Dispatch->>Rank: "Empty tensors (numel = 0, data = null)"
  Rank->>Combine: Empty expert output
  Combine-->>Router: Reconstructed local-token result
  Router->>Rank: Backward gradients
  Rank-->>Router: Completed backward pass

Loading

_{Reviews (7): Last reviewed commit: "Merge branch 'main' into allow_ep_rank_n..." | Re-trigger Greptile}

[phu0ngng]

Hold this PR until the NCCL-EP fix is upstreamed and submodule dependency is bumped.

[YangFei1990]

/te-ci L1

[phu0ngng]

Pipeline 61101023 passed except L1_pytorch_distributed_unittest--B200_8GPU is out of time due to deprecated messages, which is unrelated to this PR.