Always fully qualify types for RBIs and avoid using attributes

[vinistock]

There were a couple of things I missed in #1716.

Tapioca always uses compact class/module style (e.g.: class Foo::Bar) and never nests definitions inside a namespace. Because of that, all of the types we use must be fully qualified or else they will cause a typechecking error.

For example

class Prism::CallNode
  sig { returns(T.nilable(Location)) } # => Location does not exist. We need to use Prism::Location
  def message_loc; end
end

Also, Tapioca won't properly mix the signatures if we use attr_reader. We should always use regular method definitions in RBIs.

I tested this on Ruby LSP and the RBI generated after these changes is fully valid.

[kddnewton]

@vinistock do you want to un-nest this? i.e., make it class Prism::ParseResult`?

[vinistock]

Yes, good point. It'll make it easier to spot future issues.

I also removed the # typed: true comments from the top of the files, because Tapioca adds those automatically - and we were ending up with duplicates.

[paracycle]

Also, Tapioca won't properly mix the signatures if we use attr_reader. We should always use regular method definitions in RBIs.

I think this is actually a bug in RBI wrt its logic when doing a merge. It should treat attr_xxx defined methods and readers/writers on the same footing and merge their signatures properly.

[vinistock]

That's a good point. I created Shopify/tapioca#1709 to investigate.