GraphStage rewrite

[gafiatulin]

Fixes #21

[coveralls]

Coverage decreased (-29.4%) to 50.0% when pulling 2ad93d2 on gafiatulin:graph-stage into e209c8f on 2gis:master.

[savulchik]

@gafiatulin thanks for contributing! I will give you feedback soon)

[coveralls]

Coverage decreased (-30.7%) to 48.649% when pulling 57752b3 on gafiatulin:graph-stage into e209c8f on 2gis:master.

[savulchik]

@gafiatulin I'd like to cherry-pick library updates and cross building into master in order to release 0.3.0 version of the library having ActorPublisher implementation. After that I'd ask you to rebase GraphStage implementation on updated master.

BTW, don't you mind to split the latest commit into another PR? I'd like to discuss the approach of using typeclasses like Unmarshaller[A] for abstracting over return type.

[savulchik]

@gafiatulin I released 0.3.0 version, now you may rebase your branch on master

[coveralls]

Coverage decreased (-29.4%) to 50.0% when pulling 171355a on gafiatulin:graph-stage into cab79eb on 2gis:master.

[gafiatulin]

@savulchik I rebased on current master.
Also dropped polymorphic source commit. Going to open new issue concerning that in a minute.

[gafiatulin]

#23 is what I was looking for in the first place with that polymorphic source commit. I did it that way because it requires changes in interactions with underlying jeromq (ZMsg.recvMsg(s) vs s.recv).

Not really sure about unmarshalling being done in source. Sources shouldn't be concerned with deserializing of incoming data into domain entities. Especially not with my initial approach.