Keep the .cell wrapper around knitr floats so Typst doesn't double-wrap tables

[gordonwoodhull]

A captioned table produced by a knitr chunk (for example knitr::kable() with #| label: tbl-… and #| tbl-cap:) rendered to Typst as

#figure([
#block[
#table(

The extra #block[ between #figure([ and #table( is the shape Quarto's issue 10086 fix removes.

The cause was in q2's float desugaring. A .cell #tbl-… div was turned into a FloatRefTarget with the .cell div discarded. Quarto's Typst post filter only marks cell-output-display as scaffolding, so it disappears from the output, when it sits under a .cell parent. Without that parent, the div reached Pandoc's Typst writer and came out as a block.

q2 now keeps the .cell div around the float and leaves only the cell-output-display divs inside it, as Quarto's parsefiguredivs.lua does for non-layout cells. Layout cells and cells with no output div are unchanged.

The same knitr table now produces #figure([ followed directly by #table(, matching Quarto. typst/tbl-align-issue10086.qmd is ported from Quarto's smoke-all corpus with its two assertions. The first requires that shape, and the second rejects a block wrapped around the whole figure.

q2's output for this fixture was centered before and after, so the rendered PDF is unchanged. The fix guards against the double-wrapped #block[ #figure([ #block[ shape, which left-aligns the figure in Quarto when the scaffold rule fails to match. Because the .cell div is back, that shape would now reappear if the scaffold rule stopped matching. The second assertion covers that case.

Not addressed: for echo: true chunks using the #| label: shorthand, the code listing is still placed inside the figure, where Quarto puts it above.

[posit-snyk-bot]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

[gordonwoodhull]

(I honestly kind of hate the div shuffling and wish it were better in q2. Hard for me to merge this because it's a test for a regression in Q1, but there doesn't seem to be any corresponding bug to regress in q2. I'll probably let this draft PR decay and then close it, unless we find we need it later.)