Enable packing for all-gather CP

[huytransformer]

Description

Previously, packed CP training was rejected unless context_parallel_strategy="ring". The all-gather CP already has segment IDs and positions via the packed data, and TPU Flash attention already uses Q sharded over context with K/V gathered for attention.

b/525146921

Tests

Ran TPU unit test:

pytest tests/unit/attention_test.py::AttentionTest::test_tpu_flash_attention_packed_all_gather_context_parallel

Checklist

• 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.

[codecov]

Codecov Report

❌ Patch coverage is 20.00000% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/maxtext/utils/train_utils.py	20.00%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

[richjames0]

lgtm with nit