feat: implement URLPattern.generate() (tentative spec)

[chad-loder]

Summary

• Implements URLPattern.generate(component, groups) — the last unimplemented WHATWG URLPattern method (tentative spec).
• 19 / 19 upstream WPT cases (urlpattern-generate.tentative.any.js) now pass; previously skipped behind WHATWG_URLPATTERN_RUN_TENTATIVE=1.
• Test count: 580 → 599 (+19).
• Auxiliary-suites conformance: 84 / 84 → 103 / 103.

What generate() does

Reverses an exec(): given a component name and a mapping of named-group values, emit the canonical-form URL component string that fed back through exec() would yield the same groups.

pat = URLPattern({"pathname": "/users/:id"})
pat.generate("pathname", {"id": "42"})       # '/users/42'
pat.generate("pathname", {"id": "🍅"})       # '/users/%F0%9F%8D%85'
pat.generate("pathname", {"id": "bar/baz"})  # TypeError — would violate :id constraint

Useful for URL builders that want to round-trip values through the same pattern they match against.

Algorithm

Single pass over the per-component parsed part-list (already built at construction time). For each part:

• FIXED_TEXT with no modifier → emit the canonical value verbatim.
• Any part with a modifier (?, *, +) → TypeError (not uniquely reversible).
• Standalone full-wildcard or anonymous regex group → TypeError (no named group to substitute into).
• Missing required group → TypeError.
• Otherwise: encode the supplied value via the component's encoding callback, validate the encoded value satisfies the part's own constraint (e.g. segment-wildcard rejects path delimiters), then emit prefix + encoded + suffix.

Per-part validators are compiled lazily — generate() is a cold path; paying compile cost on every constructor for a method most callers never reach is not worth the speedup.

Implementation impact on the hot path

Zero. Three small data fields are added to _ComponentMatcher (parts, encoder, segment_wildcard_regex) — all populated from values that were already being computed at compile time. test() and exec() are untouched.

Documentation updates

• SPEC_DEVIATIONS.md — flipped generate() from "Not implemented; 19 cases xfail" to "Implemented; 19 / 19 WPT cases pass".
• README.md — auxiliary-suites badge 84/84 → 103/103; tentative-API badge tracked → implemented; WPT runners table now shows 19 / 19 for the generate suite; API surface table marks generate() as ✓ Implemented; tentative-not-implemented legend entry removed.
• docs/wpt-compliance.md — regenerated. All 19 generate cases flip from skip to pass. Footer legend simplified.
• scripts/generate_compliance_report.py — _run_generate_case now actually exercises the implementation (was hard-coded to skip).
• tests/test_wpt_generate.py — dropped the WHATWG_URLPATTERN_RUN_TENTATIVE=1 env-var gate; suite runs by default.

Test plan

• uv run pytest tests/test_wpt_generate.py -v — 19 / 19 pass
• uv run pytest -q — 599 passed, 0 skipped (was 580 + 19 skipped)
• just lint — all green (ruff, mypy, pyright, ty, semgrep, shellcheck, rumdl, codespell, interrogate, validate-pyproject)
• just compliance-report — regenerates docs/wpt-compliance.md as 469 cases, 0 failing
• Signed commit (ED25519)

Closes the v0.2.0 roadmap.