Fix bug in untypeast/pprintast for value bindings with polymorphic type annotations

[ccasin]

It is expected that you can get back a valid source program if you typecheck a program, then run untypeast on the resulting typedtree, then print it with pprintast (see tests of this in testsuite/tests/compiler-libs/test_untypeast.ml). However, this does not work for the following expression:

let foo : 'a. 'a -> 'a = fun x -> x in foo

The issue is that untypeast turns the pattern in the let binding into a Ppat_constraint node, rather than putting the type annotation in the pvb_constraint field. As a result, this invalid syntax is printed:

let (foo : 'a . 'a -> 'a) = fun x -> x in foo

This PR has two commits. The first demonstrates this issue by adding this testcase to test_untypeast.ml. The second fixes it, by simply having untypeast check whether the pattern in a value binding is a Ppat_constraint, and moving the type annotation to the pvb_constraint field in that case.

[Octachron]

I would suggest to apply this transformation only in the Ptyp_poly case to make the parsetree -> typedtree -> parsetree transformation idempotent. Otherwise, we are duplicating the constraint on the expression side.

[ccasin]

Makes sense - I've specialized the case to apply only when cty is a Ptyp_poly.

[gasche]

The parser also supports let foo : type a . a -> a = fun x -> x, does this work as expected?

[Octachron]

This syntax is elaborated in Typecore to

let (foo: 'a. 'a -> 'a) = fun (type a) -> ( (fun x -> x) : 'a -> 'a)

and it is thus is similar to the first form in the typedtree.

[ccasin]

I've added this example as a test case. It prints as @Octachron predicts.

[Octachron]

Thanks for the fix!