Cold-start-recommender

* Introduced the Data Abstraction Layer
* It's now fully in memory, with the possibility to dump everything to a DB or file
* Different instances can run

* Bug fixed

* Some debugs messages added

* Categories can now be lists (or passed as string which are json-parseable as list).
This is important for, eg, tags which can now be passed in a REST API as:

curl -X POST "http://127.0.0.1:8081/insertitem?id=Boo2&author=TheAuthor&cathegory=Horror&tags=['mio','nostro']"

* Fixed bug in recommender_api example file

* Sync categories' users and items collections in get_recommendations

Bug fixing for in-memory