✨ feat(treebuild): retarget the parser at a custom tree

[gaborbernat]

Pipelines that only need a projection of a page -- a link index, a diff tree, an extraction record, another library's nodes -- pay twice with turbohtml.parse: once for the document-sized graph of Node wrappers, and again to walk it and copy what they want out. This closes #545 by letting the same conformant parser build straight into a structure the caller controls.

turbohtml.treebuild.parse_into runs the ordinary WHATWG tree builder and drives a caller-supplied builder object -- a create_* method per node kind plus an append linking a child under its parent -- to construct the tree in one document-order pass, returning whatever the builder made its document root. No navigable Node is created and nothing is walked a second time. 🌳 The builder returns opaque handles that turbohtml only threads back into a later append, so a handle can be a node in your tree, an id, or None when you only want the calls. It is the shape Rust's html5ever exposes as its TreeSink trait and Node's parse5 as its TreeAdapter, so a parse5 adapter ports method for method.

The engine parses into the arena-backed C tree, then walks it with an iterative heap spine rather than recursion, so a deeply nested document cannot exhaust the C stack. Elements carry their namespace URI and attributes as (name, value) pairs, a <template>'s content is appended under the template handle, and a bogus <?...> construct reaches a distinct create_pi, the one place the builder is finer-grained than the parsed Document, matching the SAX stream's ProcessingInstruction. The tree construction, the walk, and the string extraction all stay in C; the builder is the single Python boundary.

The operation joins the benchmark registry next to the SAX walk, gated on the 235 kB spec, and the parse5 migration guide gains a tree-adapter section alongside the how-to, explanation, reference, and tutorial coverage.

[codspeed-hq]

Merging this PR will not alter performance

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

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
🆕	test_feature[treebuild]	N/A	24.3 ms	N/A

---

_{Comparing gaborbernat:feat/545-tree-sink (3587dda) with main (fe5325a)}

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. ↩