Refactoring ExternArgs -> SQL gen for type safety

[workingjubilee]

As noted during discussion in #683, there's currently situations where we are using purely programmatic logic with vectors of enums rather than type-driven logic, when types are fairly good at encoding logical invariants. This results in errors like the one mentioned in #640, unfortunately.

I believe we should start with tokenizing the ExternArgs more or less read off "as-is", and then use them to build struct CreateFn or similar that can only fmt::Display into a valid SQL statement. By taking in the ExternArgs "directly", and only verifying they are valid arguments in the first step, we effectively "lex" them, and then parse them in a separate step that will discern if the combined arguments can logically parse into a correct SQL statement. This is good compiler design (and PGX is more and more that: a compiler) and allows distinguishing between individually invalid and combined invalid inputs, providing more precise and useful error messages.