refactor: lexer

[psteinroe]

• adds a new tokenizer crate that turns a string into simple tokens
• adds a new lexer + lexer_codegen that uses the tokeniser to lex into a new SyntaxKind enum

the new implementation is

• much more performant (no extra string allocations, no call to C library)
• works with broken strings
• custom-made to our use-case (eg we need whitespace variants)

in a follow-up (or maybe here), we will be able to:

• parse custom parameters that popular tools use
• parse commands (lines starting with \)

we will still use libpg_query for parsing sql, but this gives us the flexibility to

• pre-process to remove unsupported stuff
• parse non-sql content (e.g. commands) via a simple custom parser

todos:

• use new lexer in splitter (to detect double newlines, we either need to make all whitespaces tokens a length of one or add a explicit "double newline" token - I think I prefer the latter because its much simpler to use and more performant)
• make sure we support all the different parameter formats popular tools use
• tests