tri_comparison_gatherer: inject css/js grammar for HTML <style>/<script> (parity with #2421)

[squid-protocol]

Summary

tests/tools/tree_sitter_accuracy_audit.py already injects the css/javascript grammar when it
hits an HTML <style> / <script> element (_html_embedded_ts_funcs, added in #2421), so its
tree-sitter side sees the real CSS/JS constructs inside. tests/tools/tri_comparison_gatherer.py
has its OWN simpler tree-sitter walk (_walk_tree_sitter) that did NOT do this — for HTML it
matched style_element / script_element against func_node_types, _get_node_name returned
None, and the tree-sitter side yielded nothing.

Consequence: cpython_jinja/layout.html:40's @media only screen { ... } inside a <style>
block — which GitGalaxy's polyglot detector correctly reports as a function media — showed up
as html/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter] (GitGalaxy alone), i.e. a
correct polyglot find looking like an over-detection.

Fix

Added the same tsaa._HTML_EMBEDDED_LANG / tsaa._html_embedded_ts_funcs branch to
_walk_tree_sitter, mirroring measure()'s own walk. media is now in GitGalaxy+tree-sitter
consensus. ctags-html still can't see it (no CSS parser — a separate documented limitation).

Found by

tri-comparison-ledger-sweep (remaining-shapes pass), 2026-08-29. Fixed in the same PR.

[squid-protocol]

Already fixed — the tsaa._HTML_EMBEDDED_LANG / tsaa._html_embedded_ts_funcs branch was added to _walk_tree_sitter in #2451 (tests/tools/tri_comparison_gatherer.py:214). The issue was filed with the fix but never closed.

Verified on main @ 5ea3eaa: cpython_jinja/layout.html:40's @media only screen { … } inside the <style> block now reads as media in GitGalaxy + tree-sitter consensus (ctags-html still can't see it — no CSS parser, a separate documented limitation). The html/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter] false-over-detection shape is gone.