Fix THD RoPE offsets for local packed shards

[kaimo455]

What does this PR do ?

Preserves THD RoPE positions for local packed shards by honoring caller-provided offsets when freqs only contains max-sequence positions.

It also adds unit regressions for the local packed-shard offset path, the fused THD fallback path when offsets are provided, and the conversion of local offset scalars to Python int values before slicing.

Validation run for this PR:

• python -m py_compile megatron/core/models/common/embeddings/rope_utils.py tests/unit_tests/transformer/test_rope.py
• PYTHONPATH=. uv run --no-project --isolated --with torch --with numpy --with packaging --with click --with requests --with pyyaml --with pytest==8.3.5 python -m pytest tests/unit_tests/transformer/test_rope.py -k 'python_ints or offsets or fusion_when_offsets' -q
• PYTHONPATH=. uv run --no-project --isolated --with black==24.4.2 --with isort==5.13.2 --with pylint==3.2.6 --with 'ruff~=0.9.0' --with mypy --with torch --with numpy --with packaging --with click --with requests --with pyyaml --with pytest==8.3.5 bash tools/autoformat.sh

⚠️ For major changes (either in lines of code or in its impact), please make sure to first share a design doc with the team. If you're unsure what's the best way to do so, contact the @mcore-oncall.

Contribution process

Pre-checks

• I have added relevant unit tests
• I have added relevant functional tests
• I have added proper typing to my code Typing guidelines
• I have added relevant documentation
• I have run the autoformatter.sh on my PR

Code review

Feel free to message or comment the @mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!

All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.

Step 1: Mark PR as "Ready for Review"

1. When your PR is ready, click Ready for Review.
2. An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
   • Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

Step 2: Final Review

For PRs that change megatron/core, once all expert reviewers have approved, the Final Review label is applied automatically and final reviewers are assigned.

For PRs outside megatron/core, this step is skipped.

Step 3: Approved

Once all required reviewers have approved, the Approved label is applied automatically.

Merge

Any member of mcore-engineers will be able to merge your PR.

For MRs into `dev` branch

The proposed review process for `dev` branch is under active discussion.

MRs are mergable after one approval by either eharper@nvidia.com or zijiey@nvidia.com.

[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.

[janEbert]

/claude review

[claude]

Clean PR overall — one potential bug flagged inline: iterating over the offsets tensor passes 0-dim tensors to _get_thd_freqs_on_this_cp_rank which expects int. Should call .item() like CASE 1 does, especially to avoid issues with GPU tensors.

[janEbert]

/ok to test ccbd017