Add opt-in WHATWG URL and URLSearchParams

[lahma]

Part of the opt-in web API series (#3079): a full WHATWG URL Standard implementation — URL and URLSearchParams behind WebApiFeatures.Url (in Default). Not a System.Uri wrapper: the spec's basic URL parser state machine, host parsers and serializers, implemented from https://url.spec.whatwg.org/ and validated against web-platform-tests.

Conformance: the vendored WPT corpus passes with zero exclusions — all 891 parse rows of urltestdata.json and all 278 setter rows of setters_tests.json (snapshot at wpt 6745634413e4153144e23bc3e866d6b14d3e55e6, BSD-3 license included, SHA recorded beside the resources). The corpus drivers run against the parser directly (no engine per row), assert minimum row counts so an unembedded resource cannot green-wash, fail on stale exclusions, and were validated by mutation: a one-character serializer mutation fails 623 rows.

Layout: Jint/WebApi/Url/Parsing/ is a pure, engine-free parser library (no Engine/Realm/JsValue) — fetch will consume it for request-URL resolution and per-redirect-hop policy checks — with the JS binding in Jint/WebApi/Url/. Percent-encoding is one 128-entry flags table over the spec's nested encode sets with Rune-based UTF-8 (USVString semantics fall out); IPv4/IPv6 are the spec algorithms including the hex/octal/overflow rules and longest-zero-run serialization.

IDNA uses IdnMapping with an important precision: the URL Standard's own domain-to-ASCII step 4 returns an ASCII domain lowercased without consulting Unicode ToASCII at all (a web-compat rule, not an optimisation — pinned with the spec's own xn--8i7caa example), so IdnMapping is reached only for genuinely internationalized domains. Its known divergences from UTS-46-as-the-spec-wants (VerifyDnsLength not disableable, CheckHyphens, empty labels, ICU version skew, invariant-globalization mode) are documented on the class with a one-time fidelity probe (faß.de) — and on this machine's ICU, none of them cost a single WPT row. This is the same posture the engine takes toward CLDR in ECMA-402.

JS surface: constructor with base, URL.parse (null on failure), URL.canParse, all component accessors with the spec's stateOverride setter algorithms (including the fact — WPT-pinned — that a failing setter still commits the components parsed before the failure), origin with the blob: re-parse rule, live-synced searchParams (update steps in both directions), delete(name, value)/has(name, value), stable code-unit sort, size, full iteration with @@iterator === entries. URLPattern and createObjectURL are absent, not throwing, so feature detection works. Subclassing works via OrdinaryCreateFromConstructor.

One hazard found in self-review and fixed with a regression test: using var on a mutable struct parser ran Dispose on a defensive copy, double-returning a pooled buffer; replaced with explicit try/finally and a test that forces the buffer-growth path.

Verified: all-TFM build, both suites × both frameworks × both host-contract-verification configurations; the WPT corpus green as above.

🤖 Generated with Claude Code