Right now yioop iterates through postings and looks up docs at the same time on each queue_server. Each group iterator grabs about 200 docs and groups these and sends to phrase model. Typically, the phrase model only needs to return the top ten results. Ranking can be done without summary look-up. Also, after grouping a lookup might need to be done on a different queue_server. So the two things should be separated and speed optimized.

Fixed by c937a147