Variant of starWars with interface subtyping failure

[NicolasRouquette]

Demonstrates the problem (#388) on a StarWars2 unit test variant that fails.

[NicolasRouquette]

Propose a conceptually simple strategy to fix #388:

Just like InterfaceType[Ctx, Val] has a field: manualPossibleTypes,
the idea is to add an extra field: manualPossibleInterfaces.

[NicolasRouquette]

@OlegIlyenko Can you please review the logic for interface subtyping in sangria.validation.TypeComparators.isSubType?

In particular, this line:

      case (t1: InterfaceType[_, _], t2: AbstractType) ⇒ (t1.name == t2.name) || schema.isPossibleInterface(t2.name, t1)

It uses a similar criteria than what's used for the Named case:

      case (t1: Named, t2: Named) ⇒ t1.name == t2.name

My original fix had a stronger criteria:

      case (t1: InterfaceType[_, _], t2: AbstractType) ⇒ (t1 == t2) || schema.isPossibleInterface(t2.name, t1)

However, this led to a failure in the schema extension test; see:
https://travis-ci.org/sangria-graphql/sangria/jobs/407010037#L1963

Since the spec says that type references are by name (see: http://facebook.github.io/graphql/June2018/#sec-Type-References) then it should be OK to use a name equality test to confirm that two types refer to the same thing.

[NicolasRouquette]

This PR is out of scope of the June 2018 Graph QL Spec