Reviewer suggestion for non-empty queries was slow and inaccurate. Before this
change, the behavior was as following:
1. For an empty query, check the previous 25 changes of the current user,
and find all the users that reviewed his change, and those who reviewed
the most changes will be ranked higher.
2. For a non-empty query, generate n*3 (n = 6 in the frontend) account
names that start with the query. E.g, for query "a" it would generate
18 account names that start with "a", and then it would rank them based
on many things (comments, approvals, etc). This is bad, since there are
much more than 18 accounts that start with "a", so the results are not
good.
In this change, the behavior for an empty query stayed the same, and for
a non-empty query the behavior is as following:
Check the previous 25 changes of the current user and find all the
people that reviewed his change that their name or email starts with
the query. E.g, if I reviewed a change, and then the owner of the change
queries "Ga" it would find me (unless 6 other people that their name
starts with "Ga" and reviewed his change).
In case there are less than 6 people that fit the query described above,
we will also find up to 6 people that just start with the name. For "a"
it will still be quite random, but for "paiki" it likely will yield good
results even if I never reviewed this user's changes.
Therefore, users are no longer ranked by comments, owning a change, etc.
They are ranked only by reviews. It also doesn't matter which project it
is they reviewed (reviewerRanking and reviewerRankingProjectIsolation
are both irrelevant tests that were deleted).
Change-Id: I0ca718d35e52b3e0794419d1f56c431af52df0b6
d73a1eccfb