[Maya-Kid]

table::grid() reads HTML table cells with dom_query's text(), which joins text nodes end to end, so words split by a <br>, <p>, <div> or <li> run together unless the source has whitespace between them.

Problem

On dev 15fd7d5, utopia_ingest::parse("page.html", …) reads

<table><tr><th>Well</th><th>Depth<br>(m)</th><th>Readings</th></tr>
<tr><td><p>North</p><p>Field</p></td><td>1200</td><td>10<br>20</td></tr></table>

as

| Well | Depth(m) | Readings |
| --- | --- | --- |
| NorthField | 1200 | 1020 |

Two readings become one number, the failure #813 fixed in Word cells and #1042 in slide table cells. Every HTML path lifts its tables through grid(): .html files, whether uploaded, synced or fetched by a URL source (Readability's article or the raw page), feed entries and their linked articles, and MinerU table bodies.

Change

• grid() reads a cell with formatted_text(), which breaks lines at <br>, <hr>, <li> and block elements but joins inline ones (<b>9</b>50 stays 950); clean folds the breaks into single spaces.

Verification

• table::tests::a_line_break_inside_a_cell_keeps_the_words_on_either_side_apart covers <br> in a header and a value, <p>, <div>, <li> and inline <b>. Without the fix it fails, printing Depth(m), NorthField, 1020, SouthRidge and 79.
• By hand (each path ends in grid()): the table above as a short page, inside a long article Readability keeps, and through fragment_to_markdown reads 1020 on dev and 10 20 here.
• Linux x86_64 container, Rust 1.98.1, PostgreSQL 16 + pgvector 0.8.6, Poppler, DB required, at 4ef3d08: fmt and clippy clean; cargo test --locked --workspace 1,564 passed, 0 failed, 6 ignored.

Limits

• Breaks drawn only by CSS (a display:block <span>) or by tags outside dom_query's list (<center>, <dt>/<dd>) still join.
• A <br> inside a Chinese label now leaves a space (归属于母公司 所有者的净利润), as Word and slide cells already do.
• Documents already ingested keep the joined text until reprocessed.

[WaylandYang]

Thank you. Reading cells with formatted_text() is the right one-line fix: it splits at <br> and block elements and still joins inline ones, so 10<br>20 stops reading as 1020 while <b>9</b>50 stays 950. That matches what #813 and #1042 did for Word and slide cells. Merging.