Refactor queries on relations and iteration on results

First, we need to debug and refactor some places in code which include work with related models. Now SQLAlchemy does too many queries, which really slows all the process.
Second, SQLAlchemy always loads all objects from queries into memory, and this can be solved on psycopg2 by using .yield_per() with queries. Then it will use Postgresql cursor and iterate through results without huge memory consumption.