[cristianchiriac]

What this changes

Document.to_markdown() no longer breaks a pipe-table row when a body cell contains a line break (<br>, or several paragraphs in one <td>). Before, the cell was written across lines, so Markdown parsers ended the table early and put the rest of the cell and the following cells outside it. A cell's lines are now joined with a space before the row is emitted, so First<br>second renders as | First second | 10 |.

Only cells that contain a newline are touched. Single-line cells, headers (already flattened in _combine_multi_row_headers) and the escaping behaviour (#1483) are unchanged.

Fixes #1486

Verification

• Regression test for a bug is in tests/issues/regression/test_1486_markdown_multiline_cells.py, and its docstring links the issue. It covers <br>, <br/><br/> and two <p> in one cell, checking both the pipe row and what markdown-it parses as table cells (["Item", "Value", "First second", "10"]). All three fail on main ('| First\n\n\n\nsecond | 10 |') and pass with the change.
• Changelog: changelog.d/1486.fixed.md; assemble_changelog.py --check accepts it.
• I ran the fast tests matching markdown, table or document before and after the change. The only difference is the new test. tests/thirteenf/test_value_units.py::test_documented_example fails in both runs because of a Unicode error in my Windows console.
• I didn't add a ground-truth assertion against the GSBD 10-K table mentioned in the issue, because it needs a recorded cassette. The synthetic table isolates the renderer path.

[dgunning]

Thanks for this. It fixes #1486 cleanly, and the fast markdown, table and document tests pass on the branch. A few small changes before merging:

1. Normalise with ' '.join(cell.split()). The current check only joins when '\n' in cell, but it splits with splitlines(). A cell holding \r, 
 and similar, with no \n, goes through unjoined and still breaks the row. ' '.join(cell.split()) covers every whitespace character. It also matches how grouped headers in the same function already flatten text, so this function would have one rule for cell whitespace instead of three.
2. Drop the markdown_it import in the test. It isn't a declared dependency and is only installed because rich depends on it. Asserting on the rendered lines directly would work; for example, check that every table line starts and ends with | and has the expected number of cells.
3. Rename the test file to tests/issues/regression/test_issue_1486_markdown_multiline_cells.py, to match the project convention.

Not something this PR needs to fix, but related: the grid and simple table formats have the same newline-in-cell problem, and a literal | in a body cell isn't escaped in the pipe format. We'll track those separately.

#1524 also changes _render_table_pipe. The two merge cleanly today.

[cristianchiriac]

Thanks for the review! All three are in 87192ca:

1. Body cells are now normalised with ' '.join(cell.split()), for every cell, so \r, tabs and the like can't break the row either.
2. The test no longer imports markdown_it. It asserts the rendered lines directly: the header, the separator and | First second | 10 |, each starting and ending with | and having two cells.
3. It's renamed to tests/issues/regression/test_issue_1486_markdown_multiline_cells.py.

The three cases still fail on main and pass on the branch. I also re-ran the fast markdown/table/document tests, and the set of failures in my environment (optional MCP/SGML dependencies and a Windows console encoding issue) is the same with and without the change.