Fix simple grid content widths

[moreaki]

What changed

This adds a small intrinsic-width helper for grid containers used by preferred width calculations.

Before this change, min_content_width() and max_content_width() treated GridContainerBox like a regular block container. That means shrink-to-fit contexts, such as floats and flex/grid items using intrinsic size contributions, could see a multi-column grid as if its intrinsic width were just the width of one child block. In the minimal example discussed in #2354/#2145, a two-column grid containing abc def twice reported a width close to one abc def, causing each grid item to wrap to abc/def.

The new helper handles simple, non-spanning row-flow grid containers by summing per-track intrinsic contributions and gutters. It is deliberately conservative: it improves common grid content width cases without attempting to implement the complete CSS Grid intrinsic sizing algorithm.

Investigation summary

I followed the #2354 link chain and related work before making this patch:

• Support for layouts with nested CSS grids and auto-height rows #2354 was opened as a follow-up to Support for flexible row heights based on content size #2342, replacing the earlier flex/grid mixed example with a grid-only example using nested grids and auto-height rows.
• Support for flexible row heights based on content size #2342 links these separate limitations:
  • aspect-ratio not supported in CSS #1891: aspect-ratio is unsupported.
  • Add support for row-gap and gap properties in flex layout #1665: flex gap support, later handled by the flex rewrite.
  • Inline-block as child of display:flex element overflows since 54 #1652 and Incorrect margins when using flexboxes #1563: flex item sizing/margin issues, later handled by the flex rewrite.
  • Grid support enhancements #2145: the grid support tracking issue.
  • Improve flex layout #2362 / Improve flex layout #2387: the flex rewrite, now merged.
• Grid support enhancements #2145 explicitly tracks Support for layouts with nested CSS grids and auto-height rows #2354 as “auto content size for grid containers”. It also tracks nearby grid gaps including Handle grid items with intrinsic size #2644 (“grid items with intrinsic size”, eg images), grid + float = overlapping text #2647 (grid + float), Allow fragmentation in grid rows #2397/Allow page breaks in grid rows #2564 (grid row fragmentation, merged), Grid rendering issues with box-sizing: border-box #2628/Handle box-sizing: border-box in grid layout #2643 (box-sizing: border-box in grid layout, merged), and Support grid-auto-flow: column #2142/Support grid-auto-flow: column #2162 (grid-auto-flow: column, merged).
• I searched for open and closed upstream PRs around Support for layouts with nested CSS grids and auto-height rows #2354/Support for flexible row heights based on content size #2342 and the relevant grid terms. I did not find an existing open PR or submitted patch specifically for Support for layouts with nested CSS grids and auto-height rows #2354.

What this fixes

The PR fixes the maintainer’s minimal intrinsic-size example from #2354/#2145:

<div style="float: left">
  <div style="display: grid; grid: 1fr / 1fr 1fr; border: 1px solid red">
    <div>abc def</div>
    <div>abc def</div>
  </div>
</div>

With this patch, the float’s max-content width includes both grid tracks, and the two abc def items no longer wrap as if the grid were one child block wide.

The PR now also includes the exact flex-column/grid sample from the #2354 issue comment as test_grid_container_in_flex_column. On current main, that standalone sample already lays out as a stretched flex item, so this test is included as related reference coverage; the behavior-changing regression remains the shrink-to-fit grid content width case above.

What this does not fix

This is not full CSS Grid intrinsic sizing support. In particular, this PR does not claim to solve:

• the complete original image/object-fit layout in Support for layouts with nested CSS grids and auto-height rows #2354;
• grid items with intrinsic image sizes (Handle grid items with intrinsic size #2644);
• aspect-ratio support (aspect-ratio not supported in CSS #1891);
• absolutely positioned or floating grid items (grid + float = overlapping text #2647);
• spanning items in intrinsic sizing;
• complex min-content / max-content cases;
• auto-fit / auto-fill;
• auto margins for grid items;
• min-* / max-* constraints on grid items;
• the complete CSS Grid track sizing algorithm for intrinsic contributions.

The intent is to provide a small reviewable base for one narrow part of #2354: simple grid containers contributing a sensible min/max-content width in shrink-to-fit contexts.

Tests

• venv/bin/python -m ruff check weasyprint/layout/grid.py weasyprint/layout/preferred.py tests/layout/test_grid.py
• venv/bin/python -m pytest tests/layout/test_grid.py
• venv/bin/python -m pytest tests/layout/test_preferred.py tests/layout/test_flex.py
• venv/bin/python -m pytest -> 4027 passed, 40 xfailed

Addresses part of #2354.
Related to #2145, #2644, #2342, #1891, #1665, #1652, #2362, #2628, #2647.

[moreaki]

I added the exact flex-column/grid sample from the #2354 issue comment as test_grid_container_in_flex_column in commit 26e28bd.

One note for review: this standalone sample already lays out as a stretched flex item on current main in my local run, so the new test is reference coverage for the related issue comment rather than the behavior-changing regression. The regression fixed by this PR is still the shrink-to-fit intrinsic grid width case, where the grid container previously contributed only one child block width and wrapped the two abc def items.

Validation after the added test:

• venv/bin/python -m ruff check weasyprint/layout/grid.py weasyprint/layout/preferred.py tests/layout/test_grid.py
• venv/bin/python -m pytest tests/layout/test_grid.py -> 52 passed, 1 xfailed
• venv/bin/python -m pytest -> 4027 passed, 40 xfailed

[liZe]

It is deliberately conservative: it improves common grid content width cases without attempting to implement the complete CSS Grid intrinsic sizing algorithm.

In this case, it may be better to have a reliable intrinsic sizing algorithm than a workaround.