[meutex]

Description

Native Word equations (OMML) lost their structure during DOCX text extraction: 1/2 + 1/3 came out as 12+13, 2^3 as 23 and sqrt(9) as 9, because the DOCX walkers collected only the leaf text runs.

This change linearizes OMML in the shared DOCX extraction, so Immersive Reading, the reading agent and chat attachments all receive an unambiguous form:

• fractions → a/b, superscripts → a^b, subscripts → a_b
• square roots → sqrt(x), n-th roots → root(n, x)
• delimiters keep their own characters and every item: [0, 1], (a,b)
• compound operands are parenthesised once: (a+b)/(c-d), (a+b)^2
• an inline m:oMath stays in place within its paragraph; a display m:oMathPara becomes its own line

deeptutor/utils/document_images.py adds linearize_omml and uses it in extract_docx_rich. The pure-OOXML fallback in deeptutor/utils/document_extractor.py reuses the same walker, so equations survive when python-docx is unavailable. No new dependency.

With the issue's math-formula-native-only.docx, all three paths (extract_material for Immersive Reading, extract_text_from_bytes for attachments, and the OOXML fallback) now give:

分数题：计算 1/2 + 1/3。
乘方题：计算 2^3。
根式题：计算 sqrt(9)。

Paragraph order, table text and embedded-image markers are unchanged.

Not covered: n-ary operators (∑, ∫), functions (sin x) and matrices still fall back to their leaf text, as before. Happy to add a notation for them if you have a preference.

Related Issues

• Closes [Bug]:Native Word equations lose their structure during DOCX extraction #2144

Module(s) Affected

• agents
• api
• config
• core
• knowledge
• logging
• services
• tools
• utils
• web (Frontend)
• docs (Documentation)
• scripts
• tests
• Other: ...

Checklist

• This PR targets dev (not main — see the base-branch selector above).
• I have read and followed the contribution guidelines.
• My code follows the project's coding standards.
• I have run pre-commit run --all-files and fixed any issues. (Ran the Python hooks on the changed files instead: ruff check and ruff format --check with the pinned ruff 0.16.0, and scripts/check_repo_hygiene.py. All pass.)
• I have added relevant tests for my changes.
• I have updated the documentation (if necessary). (No documented behaviour changed.)
• My changes do not introduce any new security vulnerabilities.

Additional Notes

On dev with this change, pytest tests/utils tests/reading: 620 passed, 6 skipped (605 passed without it). The one failure, tests/reading/test_extension_router.py::test_async_reading_action_can_be_retried_after_timeout[True], fails the same way on dev without this change (seen on Windows).