There are two data accuracy bugs fixed here. First we address the scroll
context ttl. It was only 1m which seems to not be enough for some of our
queries and we get inconsistent results. Bump that up to 10 minutes
which is what the opensearch api docs use. Note we try to clear the
scroll context when we are done but we don't appear to have enough
permissions for that so we leave the code commented out.
Then the other issue had to do with calculating the median patchsets per
review. We need to index into the sorted list not th raw list to find
the median. Fix that.
Change-Id: Iac0df1b476bf88552e829a3d1edb41790a09d99b
7e9649038a