Avoid re-gathering vocab tiling output head in backward

[huytransformer]

Description

Start with a short description of what the PR does and how this is a change from
the past.

The rest of the description includes relevant details and context, examples:

• why is this change being made,
• the problem being solved and any relevant context,
• why this is a good solution,
• some information about the specific implementation,
• shortcomings of the solution and possible future improvements.

If the change fixes a bug or a Github issue, please include a link, e.g.,:
FIXES: b/123456
FIXES: #123456

You can also provide a comma-separated list. If you don't want to close a bug but
simply to reference it, use BUGS, e.g.:
BUGS: b/123456

Notice 1: Once all tests pass, the "pull ready" label will automatically be assigned.
This label is used for administrative purposes. Please do not add it manually.

Notice 2: For external contributions, our settings currently require an approval from a MaxText maintainer to trigger CI tests.

Tests

Please describe how you tested this change, and include any instructions and/or
commands to reproduce.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

🤖 Hi @huytransformer, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

[github-actions]

## 📋 Review Summary

This Pull Request introduces a high-impact optimization for vocabulary tiling in the NNX path. By gathering the output head parameters over both FSDP and context axes before entering the custom_vjp boundary, the backward pass can reuse the already-gathered table instead of re-gathering it for every single chunk, significantly reducing communication overhead.

🔍 General Feedback

• Excellent Performance Enhancement: Moving the all-gather operation outside the custom_vjp and including the "context" axis prevents redundant communication during both forward and backward passes. This is a very clean and highly elegant JAX SPMD design pattern.
• Import Ordering: For long-term maintainability and codebase cleanliness, consider sorting the imported symbols from maxtext.utils.sharding alphabetically at the top of the file.
• Syntactic Correctness: The replacement of all_gather_over_fsdp with get_physical_spec_without_axes and maybe_shard_with_name is implemented with correct parameter arguments and aligns perfectly with existing vocabulary tiling patterns.

[github-actions]

🟡 **Testing Coverage:** While the legacy Linen path (`VocabTilingTest`) includes a `test_vocab_tiling_gradient_context_parallelism` test case, there is currently no equivalent test for context parallelism in the new NNX path (`VocabTilingNNXTest`).

To ensure the new context gathering logic on the NNX path is fully covered and to prevent future regressions, consider adding an equivalent context parallelism test (e.g., test_nnx_vocab_tiling_gradient_context_parallelism) in tests/unit/tiling_test.py.