Optimize the priority queue when selecting neighbor points

[miojizzy]

Modify

1. Modify function getNeighborsByHeuristic2
   In this function, first of all, pop all items from top_candidates one by one will generate an ordered sequence, if queue_closest is only aim to get nearest from candidates, traversal this sequence reverse order also can do it. Its no need to build another priority queue with negative distance(because of results will put in top_candidates to return, pop all items is necessary). And also delete dist in return_list (it's not use).
2. Add a simple benchmark test in /benchmark/cpp/, it will build two binary, one is use local hnswlib, another is use origin master branch hnswlib, run them seperently can get performance compair results.

Benifit

My performance test shows it will take 5%-10% time benifit during index building.
Because most of time spent on distance calculate, more data, less time benifit

dataset	avg time ns	git commit	percent
500	30833859	b125ce8	88.2
500	34959886	origin develop	100
5000	912679244	b125ce8	94.4
5000	966823741	origin develop	100
50000	16780000000	b125ce8	95.7
50000	17518000000	origin develop	100

[jianshu93]

any data on the improvement？

[miojizzy]

any data on the improvement？

no actual data yet, it is still under testing and will be released after a while