[C/PyTorch] Fixed incorrect use of torch.distributed.new_group() when creating intra-node group in initialize_ub()

[denera]

Description

#994 reported that initialize_ub() tries to create the intra-node process group with only the node ranks calling into torch.distributed.new_group() even though PyTorch requires all ranks to call into this even if they're not part of the new group.

This PR fixes the issue by creating the intra-node group via torch.distributed.new_subgroups_by_enumeration() instead. It also expands the unit test coverage over TE layers with comm+GEMM overlap and previously untested atomic GEMM overlaps.

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 refractor

Changes

• Re-implemented Userbuffers intra-node process group with torch.distributed.new_subgroups_by_enumeration().
• Added new unit tests for TE modules with comm+GEMM overlap.
• Fixed a GEMM FP8 output type mistake in comm+GEMM overlap algorithms.
• Updated existing comm+GEMM overlap tests to cover missing atomic GEMM overlaps and GEMM FP8 outputs.

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

[timmoon10]

Seems reasonable. It looks like some of these changes are also in #1067, so it might be better to merge that first.

[denera]

Seems reasonable. It looks like some of these changes are also in #1067, so it might be better to merge that first.

#1067 needs a bit more attention to validate with MLPerf benchmarks, so I'll merge this first and then rebase 1067.