✨ feat(tree): add Node.prune to keep a CSS selector

[gaborbernat]

What

Adds Node.prune(selector), the post-parse equivalent of BeautifulSoup's SoupStrainer: parse the whole document the WHATWG way, then keep only the descendants matching a CSS selector, together with their ancestors up to the node it is called on and the whole subtree under each match, removing everything else in place. It returns the node, so it chains off parse:

doc = turbohtml.parse(big_html).prune("article")

A selector that matches nothing empties the subtree.

Design

• All logic in C, in node_prune next to node_css_closest in tree_type.c, reusing the existing selector engine (selector.h) and th_node_remove.
• Free-threading safe under one per-tree critical section: pass 1 matches the selector and snapshots every match plus its ancestor chain (a regex/string filter can call back into Python, so no edit runs here); pass 2 is pure C and removes the un-kept nodes, so no structural pointer is dereferenced across a Python call or after a removal rewired it.
• Reuses the Bloom selector, interned atoms, and arena; the only allocation is one growable keep buffer.

Surface

• New Node.prune C method, registered on the shared node_methods table.
• Typed stub in _html.pyi: def prune(self, selector: str, /) -> Node: ...

Docs / tests

• How-to section, an explanation paragraph, the reference (autodoc), and a BeautifulSoup SoupStrainer -> prune migration row plus the updated omissions note.
• tests/test_tree_prune.py (12 cases) and a concurrency case in tests/test_tree_freethread.py.
• Changelog fragment docs/changelog/252.feature.rst.

closes #252