fix(serializer): resolve page-break markers when serializing a single node

[Anai-Guo]

Fixes #714.

The bug

DocSerializer.serialize() emits an internal sentinel for each page break
(#_#_DOCLING_DOC_PAGE_BREAK_<prev>_<next>_#_#, common.py:481), and every serializer replaces it in
serialize_doc(). But serialize_doc() is only reached from the whole-document path, so a page break
nested inside a group leaked the raw sentinel out of the single-node entry point — and into
HierarchicalChunker, which builds chunk.text from serialize(item=...):

whole document: '- item one\n<!-- page break -->\n- item two'
single group  : '- item one\n#_#_DOCLING_DOC_PAGE_BREAK_1_2_#_#\n- item two'   # before
chunk         : '- item one\n#_#_DOCLING_DOC_PAGE_BREAK_1_2_#_#\n- item two'   # before

The fix

A new DocSerializer._replace_page_breaks() hook, called from serialize() as well. The base
implementation returns the text unchanged; Markdown, DocTags and LaTeX override it with the
context-free replacement they already performed in serialize_doc(), which now just calls the hook.

Two details keep the existing behaviour intact:

• HTML is deliberately left on the base no-op. HTMLOutputStyle.SPLIT_PAGE needs prev_page /
  next_page and each marker's offset to slice the body into pages, so it can only resolve them once
  the whole document is available. Its sentinel survives to serialize_doc() exactly as before.
• serialize() skips the substitution when the node being serialized is itself the page break.
  Markers between two top-level items therefore still reach serialize_doc() as their own part. That
  matters for page_break_placeholder="": resolving those markers early would make them empty parts,
  which the if p.text join filters drop, silently removing the blank separators that
  test_md_cross_page_list_page_break_empty pins down.

LaTeX's serialize_doc() resolves markers from self.params.merge_with_patch(patch=kwargs) while
requires_page_break() reads self.params; the override follows requires_page_break() and uses
self.params. A per-call page_break_command override is already only half-honoured today (whether
breaks are emitted at all is decided from self.params), so I left that as-is rather than change it here.

Verification

• Three regression tests in test/test_serialization.py (Markdown, DocTags, LaTeX) serialize
  #/groups/2 of activities.json — a list spanning pages 1–2 — on its own. All three fail on main
  with the raw sentinel and pass here.
• No whole-document output changes: I exported all 38 documents under test/data/doc/ as Markdown
  (with page_break_placeholder set and empty), DocTags, LaTeX and HTML SPLIT_PAGE, before and
  after. All 190 outputs are byte-identical.
• pytest test/ — same 9 pre-existing failures as main in my environment (test_doctags_load,
  test_docling_doc; unrelated to serialization), 372 → 375 passed, the 3 being the new tests.
• ruff check and ruff format --check clean with the repo config.

🤖 Generated with Claude Code

[github-actions]

✅ DCO Check Passed

Thanks @Anai-Guo, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🔴 1 of 2 protections blocking · waiting on 👀 reviews

	Protection	Waiting on
🔴	Require two reviewer for test updates	👀 reviews
🟢	Enforce conventional commit	—

🔴 Require two reviewer for test updates

Waiting for

• #approved-reviews-by >= 2

This rule is failing.

When test data is updated, we require two reviewers

• #approved-reviews-by >= 2

---

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)(?:\(.+\))?(!)?: