Add linear algebra primitives to the API

[aalexandrov]

The DataBag type currently offers only an abstraction for parallel collections (DataBag[A]). While this allows expression of basic dataflows, certain classes of algorithms (for example, machine learning) might benefit from a direct linear algebra abstraction.

As a first part of this effort, I suggest to define two core API traits:

• Matrix[A : Numeric],
• Vector[A : Numeric],

with empty methods (def op(...): T = ???).

We can use the approach suggested by @joroKr21 in #50 and implement the operators as macros.

@fschueler, @FelixNeutatz: do you want to take a stab on this with me next week?