Spanda source tokenizer: Token, TokenType, tokenize(), unit suffix lexemes, and reserved_keywords().
Phase 4 compiler split crate — depends on spanda-ast for UnitKind mapping. Returns LexerError diagnostics; spanda-core maps them to SpandaError.