range_search_2d is faster now for counting queries

[serkazi]

When the weight range [a,b] is entirely within the query range [p,q], it is not necessary to go all the way down to $\lg{\sigma}$ levels inside _range_search_2d -- the 3-line additional check speeds up counting (of course, not reporting -- for reporting we need to recover the weights bottom-up) queries few times, on reasonably large datasets.