This gets us 37 new subtest passes in css/css-values, and 13 passes in
our other in-tree tests (and probably some random other ones!)
As noted in comments, a few parts of this algorithm have ad-hoc
behaviour to handle some issues in the spec.
The spec wants these keywords to appear in a particular order when
serialized, so let's just put them in that order during parsing.
This also fixes a bug where we didn't reject `font-variant-east-asian`
that contains `normal` alongside another value.
Also, rather than always parsing them as a StyleValueList, parse single
values on their own, and then support that in the to_font_variant_foo()
methods.
Without this, we'd happily parse `font-variant-caps: small-caps potato`
as just `small-caps` and ignore the fact that unused tokens were left
over.
This fix gets us some WPT subtest passes, and removes the need for a
bespoke parsing function for font-variant-caps.
Without this, getting a property's value from `element.style.foo` would
fail if `foo` is a shorthand property which has a longhand that is also
a shorthand. For example, `border` expands to `border-width` which
expands to `border-top-width`.
This is because we used `property()` to get a longhand's value, but this
returns nothing if the property is a shorthand.
This commit solves that by moving most of get_property_value() into a
separate method that returns a StyleProperty instead of a String, and
which calls itself recursively for shorthands. Also move the manual
shorthand construction out of ResolvedCSSStyleDeclaration so that all
CSSStyleDeclarations can use it.
