Method to speed up fulltext search indexes

[michwill]

Since all text ranking is approximate anyway, we can save something like int(log(rank)/log(k)) per word (where k is constant). And we can have, say, 20-50 of these "rank buckets". So, word rank would be OOBTree(logranks -> IITreeSet(docids)). This is needed to make indexing new documents faster. It is especially relevant when working with backups: much fewer amount of info should be read from the disk when indexing a bulk of new documents in this case