Type#contains is broken for hk types

[retronym]

I tried to fix this in one incarnation of (scala/scala#5263)

But this changed behaviour:

• Existential extrapolation was fixed, which changes overload resolution, requiring (minor) changes to scalaz.
• packedType now fails to determine that a captured existential skolem originated from outside the packing scope. The status quo depends on the fact that TypeRef#normalize caches (removing that cache would change behaviour). I was able to restore current behaviour by using the old implementation of contains in one spot in packedType, which restored compatibility with slick.
• Specs failed with an as-yet-undiagnosed problem.

common/src/main/scala/org/specs2/control/package.scala:44: value >> is not a member of org.specs2.control.package.Action[Unit]
   warn(message) >> ActionT.fail[IO, Logs, Logger, A](failureMessage)
                 ^
common/src/main/scala/org/specs2/control/package.scala:55: value >> is not a member of org.specs2.control.package.Action[Unit]
   log(t.getMessage) >>
                     ^
common/src/main/scala/org/specs2/io/FileSystem.scala:47: value void is not a member of org.specs2.control.Action[Boolean]
   action.andFinally(deleteFile(path).void)
                                      ^
common/src/main/scala/org/specs2/io/FileSystem.scala:51: value void is not a member of org.specs2.control.ActionT[scalaz.effect.IO,org.specs2.control.Logs,org.specs2.control.Logger,Boolean]
   Actions.safe(path.toFile.mkdirs).void

Noting some interaction between #185 and the problem with getting different instances of captured existential skolems from as seen from that makes us rely on the caching of normalize.