Non-executable `<script type=...>` blocks: audit walk over-counted (fixed), html `func_start` over-anchors (open)

[squid-protocol]

Surfaced by a tri-comparison-ledger-sweep pass on html (ledger shapes html/function/existence/agree[tree_sitter]_vs[ctags,gitgalaxy] and .../agree[gitgalaxy]_vs[ctags,tree_sitter]).

Context

A <script> whose type attribute is anything other than a JavaScript MIME type (or module, or bare/empty) is a non-executed data block — the browser never runs it. In language-crucible/data/html this is:

• revealjs_decks/demo.html:142,252 — type="text/template" (reveal.js slide showing sample React code as a highlighted listing)
• revealjs_decks/math.html:41 — type="math/tex; mode=display" (MathJax)
• threejs_shader_examples/*.html — type="x-shader/x-vertex" / x-shader/x-fragment (GLSL source, 10 blocks)

Part 1 — audit tooling over-counted (FIXED in this PR)

tests/tools/tree_sitter_accuracy_audit.py::_html_embedded_ts_funcs (also used by tri_comparison_gatherer.py) injected the JavaScript grammar into every <script> element, checking only for src= and never type=. It parsed the two function declarations in demo.html's text/template block (Example@145, SecondExample@158) as real functions — GitGalaxy and ctags both (correctly) report nothing there, so these counted as 2 phantom recall misses against GitGalaxy (html func recall showed 94.9%).

Fix (this PR): _html_embedded_ts_funcs now skips a <script> whose type is present and not in a small _EXECUTABLE_SCRIPT_TYPES allowlist (mirrors its existing src= skip and the HTML spec's classic/module-script gate). html ts-accuracy baseline re-blessed real_functions 39 → 37; func recall → 100%. Regression test added (tests/tools/test_html_embedded_ts_funcs.py).

Part 2 — GitGalaxy html func_start over-anchors (OPEN)

gitgalaxy/standards/language_standards.py's html func_start is <(script|style)(?=[ \t\n\r\f/>]) — it anchors a function-analog on every <script>/<style> block (intended; this is the rule behind the validated 37-occurrence agree[gitgalaxy,tree_sitter]_vs[ctags] shape, where the anchors get descriptive names from real embedded CSS at-rules / JS functions).

For a non-executable <script type="text/template"> with no embedded code GitGalaxy recognises, the anchor is left with the bare name script and no corroboration — a minor over-anchor (demo.html:142, ledger shape agree[gitgalaxy]_vs[ctags,tree_sitter], 1 occurrence). On this corpus it costs GitGalaxy 1 point of html func precision (37/38 = 97.4% vs tree-sitter's 37/37).

Proposed: a symmetric type= exclusion in html's func_start (skip <script> with a non-executable type). Deferred from this PR because it's a language_standards.py change needing its own differential scan + golden-master re-bless, and must be checked against the validated 37-occurrence shape.

Acceptance

• _html_embedded_ts_funcs skips non-executable <script type> (this PR)
• html func_start skips non-executable <script type> (follow-up)

[squid-protocol]

Reopening — #2493 closed this via a 'Closes part 1 of #2492' phrasing, but part 2 is still open: the symmetric non-executable-type= exclusion for HTML's own func_start regex in gitgalaxy/standards/language_standards.py. Picking that up now.