Reject lexer errors during parse

[StantonMatt]

Closes #63.

parse("value =< 42") was succeeding because Chevrotain reported = as a lexer error, but the parser only received the remaining valid tokens and parsed the expression as value < 42. This now returns a parse failure whenever tokenization reports invalid characters, so lexer errors cannot be silently dropped before parsing.

I also updated the existing newline-in-string expectations to assert the new lexer-error message instead of the previous parser EOF message.

Validation:

• pnpm exec vitest src/spec/index.spec.ts --run failed before the fix with the new regression and passed after the fix
• pnpm test
• pnpm lint
• pnpm type-check
• pnpm build
• git diff --check
• review-fix-loop clean: no actionable correctness issues