Fix ord on variants conflicting with Result

[sim642]

Closes #254. Besides Error also applies to Ok, coming from Ppx_deriving_runtime.

Conceptually the fix is simple: just add a type constraint to the to_int function generated into the wildcard_case of variant comparison. This forces type-directed construction resolution to correctly use the constructor from our type instead of Result.

The tricky part was not breaking other tests, where a new variant type is defined with the same name as a standard one, e.g. bool. In that case the to_int type constraint inside the comparator must still refer to the outer bool type being declared, not the one from Ppx_deriving_runtime, which happens to be opened at that point due to sanitize.
A local module definition Ppx_deriving_ord_helper is used to capture the type being declared outside of sanitize, such that inside it, we can refer to the correct bool. Hopefully such type-only local module doesn't have any runtime cost (?), otherwise I'll have to make it conditional to only happen for variant types (currently it's generated unconditionally and otherwise just unused).

[sim642]

I'm guessing my fix only works with OCaml 4.11 and up due to the following point in OCaml 4.11.0 release notes:

• #6673, #1132, #9617: Relax the handling of explicit polymorphic types. This improves error messages in some polymorphic recursive definition, and requires less polymorphic annotations in some cases of mutually-recursive definitions involving polymorphic recursion. (Leo White, review by Jacques Garrigue and Gabriel Scherer)

I tried slapping Ppx_deriving.strong_type_of_type around the to_int type constraint, but that revealed a very confusing error message from the typechecker: ocaml/ocaml#11129.

Changing that to avoid the Ptyp_poly when there are no free variables leads to a problem with the actually polymorphic cases:

Error: This expression should not be a function, the expected type is
       'a. [ `Cons of 'a | `Nil ] -> Ppx_deriving_runtime.int

I hopefully managed to make it all work by also inserting Pexp_newtypes in the to_int expression, but it's surprisingly involved and I'll have to clean it up before I push it here.