AppliedTypes returned by baseType shouldn't have type lambdas as constructors

[dwijnand]

AppliedTypes returned by baseType shouldn't have type lambdas as constructors

In tests/pos/argDenot-alpakka.min.scala, we want

`(([O] =>> Foo[O, S]) @uV)[Int]`.baseType(`Foo`)

to return Foo[Int] rather than an applied type lambda, so that we can rely on
the invariant that the type arguments of the type returned by baseType(cls)
correspond to the type parameter of cls which NamedType#argDenot is implicitly
relying on.

This could be achieved by removing the initial if branch from the AppliedType
case in baseTypeOf, since the recursive fallback will always work, but it makes
sense to keep a special case for performance, so we just explicitly add as a
condition to the fast-path that the type constructor of the AppliedType can't be
a lambda.

[smarter]

I went with fixing the input before baseType - the alternative would be
to make baseType do the same type constructor dealiasWidening.

Does anything go wrong if we change baseType instead? Intuitively I would expect it to always return an AppliedType(Foo, ...).

[dwijnand]

Let's see what all of CI says. WDYT of doing it like this?

[smarter]

WDYT of doing it like this?

I think this is the right place to fix this, but I wasn't a fan of having a slightly different typeSymbol, so I pushed a commit with a different approach where I explicitly check the shape of the type constructor.

[smarter]

@dwijnand if this looks good to you I can squash and merge.

[dwijnand]

Can't approve my own PR. Feel free to force push if you want and merge. Thanks, Guillaume.