audit: bison review & fixes

[theodevelop]

Type of change

• Bug fix

What does this PR do?

Two bugs found during a proactive audit of bisonParser.ts, analogous to the Flex audit in #35.

audit-E — uppercase-only regex in %left/%right/%nonassoc parsing
[A-Z_][A-Z0-9_]* in bisonParser.ts:169 silently dropped lowercase/mixed-case
tokens (e.g. kPLUS, kMINUS, tTOKEN) from doc.precedence. These tokens
were then falsely reported as bison/undeclared-token (or missed by the
shift/reduce heuristic). Fixed to [A-Za-z_][A-Za-z0-9_]*.

audit-F — [^}]* in extractDollarRefs misses $N after nested sub-blocks
/\{([^}]*)\}/g stops at the first }, so a $N reference appearing after
a nested { ... } block inside the same action (e.g.
{ if (cond) { log(); } $$ = $5; }) was not detected, producing a false
negative for bison/out-of-bounds. Rewrote extractDollarRefs with a
brace-depth scanner. Added removeBalancedBraces() helper and applied it to
extractSymbols, getFirstSymbol, and extractRuleReferences for consistency.

Related issue

No dedicated issue — found during internal audit.

How to test manually

1. Open a .y file with %left kPLUS kMINUS (lowercase tokens) — no false
   bison/undeclared-token should appear for kPLUS/kMINUS.
2. Write a rule with an action like A { if (1) { int x = 0; } $$ = $5; } —
   $5 should be flagged as out-of-bounds (1 symbol + 1 action = 2 positions).

Checklist

• npm run compile passes with no new errors
• Tests added or updated — 2 new assertions in tests/test-diagnostic-codes.ts (196 total, 0 failed)
• Manual test done in VS Code
• CHANGELOG.md updated
• No unintended files staged