Assertions on precise generic data representation.

I like being real picky with my generic code, disallowing misuse at the type level. However, this makes it less flexible overall, and is a large chunk of the code I have to write over and over again.

I mainly care about sanity checks, along the lines of "if the generic representations looks like this, type error out". So they don't make any term-level changes.

So, instead of hiding these in generic type class instances, I put them in type family equations. Now we can turn these checks on and off -- and when they're on, you have to carry around that fact in your types. Fantastic!

Checks are formed as Constraints, where a failure triggers a TypeError and a success goes to the empty constraint () (() :: Constraint as well as Type).

These checks were always done on the type-level, but they were "inline" with the rest of the type class. By pulling them out, we *should* be incurring some compile-time performance penalty (albeit hopefully minor due to the simple nature of the checks), but making no change to runtime.