Stop reserving builtin module names as lexer keywords (v3.5 final item)

[particlesector]

Summary

• cube/sphere/translate/scale/etc. are no longer reserved lexer keywords. Matching real OpenSCAD (where builtin modules and variables live in separate namespaces), a script can now use one of those names as an ordinary variable, module/function parameter, or for/let loop variable — e.g. cube = 5; sphere(cube); or module box(scale) { cylinder(h=scale, r=1); } now parse correctly.
• The Lexer's keyword table is trimmed to genuine OpenSCAD grammar keywords only (if/else/for/each/module/function/let/include/use/undef/true/false). Builtin construct names lex as plain Ident.
• The Parser gained a kBuiltinNodeNames lookup used only at statement-start call position (name(...)) to route to the existing primitive/boolean/transform/extrusion parsing helpers by name instead of by dedicated token kind. Everywhere else (assignments, parameter lists, loop variables, named call arguments) these names now behave as ordinary identifiers, which incidentally also fixes named module/function-call arguments like scale=4 that previously failed to parse.

This was the one item explicitly deferred from the v3.5 completeness audit (docs/roadmap.md) and closes out feature completeness against OpenSCAD's builtin/keyword surface.

Test plan

• Full test suite (chiselcad_tests, Lexer/Parser/Interpreter/CsgEvaluator/loader suites) built standalone with g++ + system Catch2 (no Vulkan/Manifold toolchain in this sandbox) — 2915 assertions / 539 cases pass.
• Added 6 new parser test cases covering builtin names used as top-level variables, transform arguments, module parameters, for/let loop variables, and named call arguments.
• Updated existing lexer tests that asserted the now-removed keyword token kinds.
• Manual end-to-end parse check of a script mixing cube, scale, and rotate as both variables/parameters and builtin calls in the same file — parses with zero errors.

---

Generated by Claude Code

[particlesector]

Reviewed. This correctly closes the last v3.5 gap: builtin names (cube, scale, etc.) are dropped from the lexer's keyword table and instead resolved by name only at statement-start call position via kBuiltinNodeNames in the Parser. Verified the logic handles modifiers (#cube();), named module-call args (scale=4), and that no other code depends on the Lexer tagging builtins specially — the TokenKind values are now purely internal Parser tags. Test coverage (new parser cases + updated lexer expectations) looks appropriately targeted, and CI is green. One minor robustness nit left inline on the fallthrough behavior if the builtin map/switch ever diverge; not blocking.

---

Generated by Claude Code