✨ feat(clean): add XML/XHTML output to the sanitizer

[gaborbernat]

Sanitizing is not unique to HTML: an XHTML dialect such as Reportlab's RML or an ePub content document rejects HTML's bare <br>, so a bleach-based cleaner has to patch its output with a brittle .replace("<br>", "<br/>") or reparse and reserialize the whole tree a second time. Policy now takes an xml flag that removes that step: the sanitizer serializes its cleaned tree as well-formed XML/XHTML instead of HTML, so sanitize("<p>one<br>two</p>", Policy(tags=frozenset({"p", "br"}), xml=True)) returns <p>one<br/>two</p>. This closes #565 and clones DOMPurify's PARSER_MEDIA_TYPE: 'application/xhtml+xml'.

The allowlist walk and the non-configurable safety baseline are untouched, so an XML-mode policy is exactly as safe as its HTML-mode twin, and only the final serialization changes. 🔒 Because the sanitizer keeps its parse-once, serialize-once shape with no reparse round trip, there is no new opening for a mutation-XSS vector to re-enter. The well-formedness pass is opt-in on the serializer, carried by a well_formed serialize option that only the sanitizer's inner_xml turns on: on that path a comment's -- and trailing - gain a space, a character outside XML's Char production (a C0 control, a lone surrogate, U+FFFE/U+FFFF) is dropped, a stored xmlns the serializer already declares for a foreign root is not duplicated, and an attribute whose name is not a valid XML name is skipped. A plain Node.serialize(Html(xml=True)) stays on the raw XML path, byte- and instruction-identical to before, so the serialize benchmarks do not move. The new Node.inner_xml property exposes the well-formed children serialization the sanitizer routes through.

Every vector in DOMPurify's own XSS corpus (its test/fixtures/expect.mjs, vendored as a pinned submodule) is run through the XML sanitizer to confirm each one serializes to output that is both inert and reparses through turbohtml.parse_xml, and an lxml differential checks the same well-formedness against an independent parser. The xml flag is off by default, so existing callers are unaffected.

[codspeed-hq]

Merging this PR will not alter performance

✅ 86 untouched benchmarks
🆕 1 new benchmark
⏩ 18 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
🆕	test_feature[sanitize-xml-spec]	N/A	12.2 ms	N/A

---

_{Comparing gaborbernat:feat/565-sanitize-xml (c082f67) with main (f73cae2)}

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. ↩