Typer#escapingRefs: don't let the types of lower bounds escape

[smarter]

In 0efa171 I changed the definition of NamedPartsAccumulator to exclude lower bounds as this is required for the implicit search, but NamedPartsAccumulator is also used by Typer#escapingRefs so in the following code:

class Foo[T]
val z = {
  class C

  ??? : Foo[_ >: C]
}

the type of z was inferred to be Foo[_ >: C] instead of Foo.

To avoid this, NamedPartsAccumulator will only exclude lower bounds if the parameter excludeLowerBounds is explicitely set to true.

No test because there is no way to detect that a type has escaped, this might be something that could be added to TreeChecker.

---

This PR also removes dead code from Typer#escapingRefs.

---

Review by @odersky .

[odersky]

In fact, given the observed failure of implicits, I would revert completely. Implicits should be searched in lower as well as upper bounds. So NamedParts can be reverted to use both bounds.

[smarter]

@odersky : That's one way to do it, but I think this is unnecessary, the failures we get for implicits are due to issues in the inference of type parameters: #739

For example, in #553 we have the type CC? >: ([+HK$0] => scala.collection.immutable.List[HK$0])[HK$0 =+ (T? >: Int)] <: ([HK$0] => <: scala.collection.TraversableOnce[HK$0]) and implicit search fail, but if #739 was fixed then CC would be instantiated to List like it is in Scala2 and the implicit search would succeed.

I have a prototype that fixes the inference of type parameters, but it uncovered some issues with Typer#ensureNoLocalRefs which I need to fix first, including the issue fixed by this pull request.

[smarter]

@odersky : could you take another look at this? I just rebased it because it turns out I need this for the sbt phase.

[odersky]

LGTM