tree_sitter_accuracy_audit _get_node_name: css keyframes_statement branch is dead code; bodyless @layer counted as a function

[squid-protocol]

Summary

tests/tools/tree_sitter_accuracy_audit.py's _get_node_name mis-reads two CSS at-rule node
shapes, making the tri-comparison audit disagree with GitGalaxy where GitGalaxy is actually
correct. Both surfaced during a tri-comparison-ledger-sweep pass on css. Same bug class as
#1313 (which fixed media_statement / supports_statement but left these two wrong).

Neither is a GitGalaxy engine defect — GitGalaxy's css func_start is correct in both cases.

Bug 1 — keyframes_statement name branch is dead code (precision)

if node.type == "keyframes_statement":
    for child in node.children:
        if child.type == "at_keyword":        # never matches in this grammar version
            return child.text.decode("utf8").lstrip("@")
    return None                               # always taken

In the pinned tree-sitter-css grammar the at-keyword child of a keyframes_statement has node
type @keyframes (or @-webkit-keyframes), not at_keyword. The loop never matches, so
_get_node_name returns None for every @keyframes node and the walk drops it. Net effect:
tree-sitter reports 0 @keyframes across the entire corpus, even though its parse tree
contains the nodes and GitGalaxy's func_start matches @keyframes fine.

Ledger shape: css/function/existence/agree[gitgalaxy]_vs[ctags,tree_sitter] (5 occ) —
tailwindcss_atrules/theme.css:443,449,457,463 (nested in a Tailwind-v4 @theme block) and
gutenberg_css_modules/button.module.css:237 (top-level). All 5 are real @keyframes rules;
GitGalaxy is right, the audit reader was blind.

Bug 2 — bodyless @layer a, b; counted as a function (recall)

_get_node_name's at_rule branch returns "layer" for both:

• @layer wp-ui-components { … } — block form, which GitGalaxy's func_start matches (its
  (?=[^{]*\{) lookahead requires a block)
• @layer wp-ui-utilities, wp-ui-components, …; — a bodyless cascade-layer ordering statement,
  which GitGalaxy deliberately does not match

GitGalaxy's css func_start counts only block-bearing at-rules — the same
"body-bearing definitions only" convention this very audit tool already applies when it drops
C/C++ forward declarations (class Foo;). The bodyless @layer a, b; is the CSS analogue. The
audit should not credit it as a real function GitGalaxy missed.

Ledger shape: css/function/existence/agree[tree_sitter]_vs[ctags,gitgalaxy] (4 occ) —
gutenberg_css_modules/{button,card,tabs}.module.css:1 and tailwindcss_atrules/index.css:1.
(3 of the 4 are additionally obscured by rank/line occurrence-alignment fuzz — GitGalaxy's count
for the block form in those files is already correct.)

Fix

_get_node_name, keyframes_statement branch: accept child types @keyframes /
@-webkit-keyframes (keep at_keyword for forward-compat), fall back to "keyframes".

_get_node_name, at_rule branch: return None when the node has no block child.

at_rule / keyframes_statement appear in no other language's NODE_MAPS entry, so the change
is css-only. After the fix, GitGalaxy and tree-sitter agree on the function count in every one of
the 21 css corpus files (zero diffs), recall_audit.py css counted-misses goes 4 → 0, and css
Func Found / Func Precision both reach 100% with no asterisk.

Fixed in the same PR as this issue, alongside the two ledger verdicts and the
docs/language_status/css.md §9 capstone.

[squid-protocol]

Fixed in #2500 (both branches — the keyframes_statement dead-code branch and the bodyless @layer over-count). css func precision + recall now 100%, no engine change.