[eXpl0it3r]

Description

Core/CssEngine was PeachPDF 0.9.6's copy of ExCSS. ExCSS itself has received 118 commits since 4.3.2 in the mean time, a good chunk of them upstreamed from PeachPDF, so this replaces the copy with ExCSS master (589afaf), kept internal and in our namespace. On top of it, only the hooks the renderer needs (Stylesheet.IsUserAgent, Length.ToPixels, the media feature helpers, CalcEvaluator, CodepointRange) and a few fixes ExCSS doesn't have stay local:

• Escaped surrogates become U+FFFD
• An alignment prefix like safe without a keyword no longer throws
• transition: none is accepted
• list-style-type: hebrew is accepted
• text-shadow takes inset and a non-negative spread distance (Text Decoration 4)
• perspective-origin: top 2px is rejected, only keyword pairs may be swapped
• Math functions are accepted in grid track sizes and basic shapes
• font-palette accepts palette-mix()

The last four are things the old copy got right and ExCSS master doesn't. They're good candidates to upstream to ExCSS, together with the first four.

ExCSS represents a few things differently, so the renderer is adapted:

• The flow-relative properties (margins, padding, inset, borders, radii, float: inline-start) are properties of their own and get mapped to the physical sides, LTR as before
• white-space and text-align are shorthands now, their longhands are combined back, and text-align-all and font-variant-caps map to the properties the box knows
• :first-child and friends come in as plain pseudo-classes and are matched by the renderer, and the s flag of attribute selectors compares case-sensitively
• A declaration with var() is applied after the cascade, and a later declaration only overrode it under the exact same name. With the flow-relative properties and the new shorthands that broke, e.g. margin-inline-start: var(--m) won over a later margin-left. A later declaration now wins for every longhand it sets, which also fixes margin: var(--m) overriding a later margin-left on master

ExCSS also accepts values that broke the rendering, because the old copy simply rejected them:

• display: flow-root and display: contents collapsed the box to 0×0. flow-root lays out as block now, contents stays ignored as before.
• rgb(calc(…)), relative colors and light-dark() ended up transparent. Colors the renderer's own parsers don't know are resolved through the engine, and light-dark() gives its light color, as long as color-scheme is ignored.
• overflow: clip was treated as a scroll container and a new formatting context.

The spec manifests are updated accordingly. Over 100 gap tests pass now and lost their [Ignore], a few more were split into the part that passes and the part that doesn't, and the recorded gaps went from 1'558 to 1'445. The four PeachPDF-only grammars that only tests used are gone, their tests now check what the parser accepts and rejects.

How to test this PR?

CI should succeed. Additionally:

• Every remaining gap test still fails without its [Ignore]
• The spec compliance check (dotnet run --project Source/Tools/HtmlRenderer.SpecCompliance -c Release -- check) reports no problems
• The image regression baselines are unchanged