✅ test(xslt): validate XSLT 1.0 against libxslt REC suite

[gaborbernat]

The XSLT 1.0 processor from #537 shipped with green coverage, but coverage proves the code runs, not that it matches the spec. This validates turbohtml.transform against an authoritative external oracle: libxslt's own XSLT 1.0 Recommendation corpus, the REC and REC2 stylesheet/source/expected-output triples the reference implementation behind lxml.etree.XSLT ships. It vendors that corpus as the pinned tests/conformance/libxslt shallow submodule and runs turbohtml over every triple, asserting byte-equal output with whitespace normalized per output method.

Of the 79 cases, 56 pass exactly. The remaining 23 exercise features turbohtml does not model, so each is xfail tagged with the spec section it needs: whitespace stripping (xsl:strip-space), xsl:namespace-alias, attribute sets, multi-level xsl:number, cdata-section-elements, extension elements, and the html method's meta injection. 📋 The corpus is a git submodule, not a runtime library, so an absent checkout is a setup error: the module raises at import telling you to run git submodule update --init, and the dedicated conformance CI job always checks it out.

The oracle surfaced four conformance bugs, all fixed here. A literal result element now copies its in-scope stylesheet namespace declarations to the output per section 7.1.1, de-duplicated against the ancestors already in scope and filtered by exclude-result-prefixes, so a prefixed or default-namespaced result element stays well-formed instead of losing its declaration; xsl:* directive attributes on a literal result element are stripped rather than emitted. xsl:number honors the separators around a format token, so format="1. " yields 1. and format="(1)" yields (1), and a value with no alphabetic or roman form falls back to decimal 🔢. A roman numeral past 4999 previously produced an unbounded run of M, a memory-exhaustion hazard, and zero or a negative produced empty output. A processing instruction now serializes with the ?> terminator under the xml output method.

Line and branch C coverage stays at 100% on the changed xslt.c and document.c. Three existing unit tests that had pinned the pre-fix behavior are corrected to the libxslt-confirmed output.

validates #537

[codspeed-hq]

Merging this PR will not alter performance

✅ 81 untouched benchmarks
⏩ 18 skipped benchmarks¹

---

_{Comparing gaborbernat:test/537-xslt-conformance (59c51d8) with main (e16e98c)}

Footnotes

1. 18 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩