feat(style): implement the CSS font shorthand

[danperks]

What & why

font was missing from the property dispatch entirely — the only "font" arm in
declaration.rs is the one that maps the legacy HTML <font> element's
attributes. So font: 10px/1 Ahem set nothing at all, and the element stayed at
the default 16px in the default family.

That matters more than a normal missing property, because font: <size>/<lh> Ahem
is the idiom WPT reaches for whenever a test needs predictable glyph metrics.
13.4% of files under css/ use the shorthand (6,574 of 49,014). Those tests
weren't failing because the feature under test was broken — they were being
measured with the wrong ruler, and a test whose reference uses the shorthand fails
even when the test itself never mentions it. On a 5,295-test sample across
css-text, css-flexbox, css-grid, css-backgrounds, css-sizing and
css-overflow:

	pass	fail	total	pass rate
test or its reference uses font:	170	1,406	1,576	10.8%
neither does	799	2,920	3,719	21.5%

1,406 of the 4,326 failures in that sample (32.5%) sat behind this.

What it does

Parses the full grammar:

[ <font-style> || <font-variant-css2> || <font-weight> || <font-stretch> ]?
<font-size> [ / <line-height> ]? <font-family>

• The optional prefixes in any order, one per category, including numeric weights.
  font-variant and font-stretch aren't modelled by ComputedStyle, but they're
  still recognised — an unrecognised token has to invalidate the whole declaration,
  so silently not knowing them would reject valid values.
• / <line-height> in all four spellings the tokenizer can hand over: 10px/1,
  10px/ 1, 10px /1, 10px / 1.
• The <system-family-name> keywords (caption, menu, …). We can't ask the
  platform what those are, so they land on the UA default — but they still reset the
  longhands, which is the part that's observable.
• inherit / unset (the font longhands all inherit, so those coincide) and
  initial.

Two behaviours worth calling out, both of which have tests:

• Size and family are required. A value that yields neither is invalid and must
  leave the cascade alone, so nothing is written to the style until the whole value
  has parsed — font: 12px can't be allowed to land the size and drop the family.
• It resets the longhands it covers. font-weight: bold; font: 20px serif ends
  up non-bold, and a line-height from an earlier declaration in the same rule
  doesn't survive. The line-height is resolved against this declaration's size,
  not the size the element had on the way in, so font: 10px/1 is 10px and not 16.

<absolute-size> keywords come along for the ride: the shorthand's <font-size>
accepts them and parse_font_size previously returned None for medium /
x-large / …, leaving the inherited size in place. They're now on the usual
medium = 16px scale.

Verified end-to-end as well as in unit tests — rendering font: 10px/1 monospace,
font: 10px monospace and font: italic bold 10px/2 monospace now produces line
boxes of 10px, 13px and 20px respectively, matching the equivalent longhands, where
before all three produced 21px (16px at the default line-height).

🤖 How this was built

• Authored primarily with an LLM (which: Claude Opus 4.5 via Cursor)

Checklist

• PR title is a Conventional Commit
• cargo test --workspace passes
• cargo fmt --all + cargo clippy clean
• WPT report (auto-commented below) shows no conformance regressions

One note on the WPT delta

Expect this to show up as a mixed delta rather than a clean win, and the
regressions are worth reading carefully — several of them are tests that were only
passing because both sides were wrong in the same way. The clearest example is
css/css-grid/grid-items/grid-item-block-axis-content-contribution-00{1,2,3}:
test and reference both say font: 10px/1 Ahem, so with the shorthand ignored both
rendered at 16px and matched. With it honoured, the reference (explicit
height: 10px) is right and the test's auto-sized grid row comes out 13px, because
grid items ignore line-height — a separate, pre-existing bug that this change
merely stops masking. I'll send that one separately.

Made with Cursor