This pull request updates the match_tags post-processing pass which calls find_matching_close_tag for every open tag, scanning forward through all remaining nodes.

With many unclosed tags (e.g. 50,000 × <div>), this is O(n²), each open tag scans to the end finding nothing.

A first pass now collects which tag names have closing tags. Open tags whose name isn't in that set skip the scan entirely, reducing the pathological case from O(n²) to O(n).

For 50,000 unclosed tags this went from ~13s down to ~0.17s.