Fix is_rec_typ deriver bug(s)

[jmid]

This is an attempt at fixing #269.
With the fix in 3c1f147 we are able to derive a straight-forward, type-correct generator for type t = Foo of t list (yay!)

let rec gen_sized n =
  QCheck.Gen.map (fun gen0 -> Foo gen0) (QCheck.Gen.list (gen_sized (n / 2)));;
let gen = QCheck.Gen.sized gen_sized

which blows the stack (less yay):

# Gen.generate1 gen;;
Stack overflow during evaluation (looping recursion?).

I think this is to be expected from the current simple, but predictable approach.
It could potentially be improved if we incorporate passing sizing explicitly to the derived list generator and thus try to reach a base-case that way.

While fixing this I noticed there was also a problem with record declarations in is_rec_constr_decl.
With the fix in daa357e for that, one can derive a functioning generator for a type like type t = Foo | Bar of { baz : t }:

# Gen.generate1 gen;;
- : t = Bar {baz = Foo}
# Gen.generate1 gen;;
- : t = Foo

Overall, this PR should leave the deriver in a slightly better shape than before.
I don't claim it to be bugfree on all language cases though - I think there are still core_type cases in ast.ml that we may not be handling correctly, e.g., for type constraints.

[jmid]

Rebased on main to reenable CI