feat(cli): add latex output format

[aron-intframe]

The LaTeX serializer has been available in docling-core since v2.54 (docling-project/docling-core#445), but the docling CLI cannot emit it. This adds latex to OutputFormat and wires it through export_documents, so docling <source> --to latex writes <name>.tex (standalone document, default serializer params). Addresses the LaTeX part of #343; mathpix-markdown-it is not covered.

Verified against docling-core's own golden files: converting test/data/doc/{2408.09869v3_enriched,activities,polymers}.json with --to latex reproduces the corresponding .gt.tex exactly (43,965 / 1,344 / 10,909 chars). Also smoke-tested end-to-end on tests/data/pdf/sources/2305.03393v1-pg9.pdf (sections, table, caption render as expected).

Known limitations, stated upfront: pictures are emitted as % image placeholders (referenced-image wiring could follow in a later PR), and convert-remote needs a service version that also knows latex.

Checklist:

• Documentation has been updated, if necessary. (docs/usage/supported_formats.md; the CLI reference is auto-generated)
• Examples have been added, if necessary. (none needed)
• Tests have been added, if necessary. (tests/test_cli.py::test_cli_exports_latex)

[github-actions]

✅ DCO Check Passed

Thanks @aron-intframe, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[PeterStaar-IBM]

lgtm!

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[aron-intframe]

Thanks for the approval. Before it lands I re-verified the branch as it stands (98847320, merge-base with main is d2283add = current main, so 0 commits behind), and I found one thing in my own PR description that needs correcting.

Correction to the PR body: "reproduces the corresponding .gt.tex exactly" is off by exactly one character. The CLI output matches docling-core's golden LaTeX files in content, but the .gt.tex files end with a final newline and the CLI does not write one. Running the CLI over the three docling-core fixtures (docling-core checked out at v2.92.0, the version resolved here) and comparing character-for-character:

$ for n in 2408.09869v3_enriched activities polymers; do \
    python -m docling.cli.main docling-core/test/data/doc/$n.json \
      --from json_docling --to latex --output out_tex; done
$ python compare.py
2408.09869v3_enriched: cli_chars=43965 gt_chars=43966 equal=False equal_ignoring_final_newline=True
activities: cli_chars=1344 gt_chars=1345 equal=False equal_ignoring_final_newline=True
polymers: cli_chars=10909 gt_chars=10910 equal=False equal_ignoring_final_newline=True

The whole delta is the trailing \n:

$ diff out_tex/activities.tex docling-core/test/data/doc/activities.gt.tex
61c61
< \end{document}
\ No newline at end of file
---
> \end{document}

I have not changed this on the branch, because it is what the CLI already does for its other text outputs — DoclingDocument.save_as_markdown ends in filename.write_text(md_out, encoding="utf-8") with no newline appended either, so .tex behaves like .md today. If you would prefer .tex files to end with a newline, say so and I will push the one-line fp.write(ser_res.text + "\n"); I did not push it unprompted since it would touch an already-approved branch.

Tests re-run on the branch:

$ python -m pytest tests/test_cli.py::test_cli_exports_latex -q
1 passed in 3.29s

$ python -m pytest tests/test_cli.py -q
2 failed, 55 passed in 4.79s

The two failures are test_cli_help and test_cli_convert, and they are environment-only: my venv installs [convert-core,extract-core,format-pdf,format-office,format-opendocument,format-web,format-latex,format-email,cli,feat-chunking] plus CPU torch, so docling_ibm_models and the service-client extra are missing. I checked out main at d2283add in the same venv and got the identical two failures, so they are not from this PR.

Not verified, stated plainly: (1) convert-remote --to latex end to end, since I have no docling-serve instance here that advertises latex — that is still the known gap in the PR body; (2) compiling an emitted .tex with a real LaTeX engine, since there is no TeX distribution in this sandbox. The --to latex output is byte-compared against docling-core's own goldens rather than typeset.