First trait for streaming testing

[juanrh]

Only support forall -> always, but allow to specify assertions at the batch level and at the individual record level

[juanrh]

• encapsulate in a trait for nice code reuse. Consider Shapeless HList https://github.com/milessabin/shapeless/wiki/Feature-overview:-shapeless-2.0.0 to simulate heterogeneously typed varargs, instead of relaying on several overload for different arities. But it is important that the user only needs to use raw tuples, use the typpe reconstruction capabitilies of .cast, or "facilities for abstracting over arity": we can ask the user to use a convention based on position of the argument, but the type should be preserved.
• create derived dstreams from the generated dstream, and assert on them, abstracting away from test multiplexion: try with a derived dstream for map(_+1) and adapt the assertions
• add some test examples: use TL generators for that?: don't forget test that ensures that batches are not mixed, following the one in the PoC for #6

[juanrh]

Covered by SharedStreamingContextBeforeAfterEach. Splitting this issue in a trait for streaming specs2 test and props for scalacheck + specs2 in streaming, covered by #15 and #4 among others