[Merged by Bors] - ref: Switch repo tests query to be a (somewhat) faster join

With tweaking random_page_cost, this returns in an acceptable period of time, albeit still slow:

 Sort  (cost=5336923.41..5337315.59 rows=156873 width=167) (actual time=15894.227..15895.152 rows=4381 loops=1)
   Sort Key: (avg(t.duration)) DESC
   Sort Method: quicksort  Memory: 1359kB
   InitPlan 1 (returns $0)
     ->  Index Scan using unq_repo_name on repository  (cost=0.27..18.77 rows=1 width=16) (actual time=0.021..0.044 rows=1 loops=1)
           Index Cond: (((owner_name)::text = 'getsentry'::text) AND ((name)::text = 'sentry'::text))
   ->  GroupAggregate  (cost=5312091.27..5316013.10 rows=156873 width=167) (actual time=13579.396..15890.345 rows=4381 loops=1)
         Group Key: t.hash, t.name
         ->  Sort  (cost=5312091.27..5312483.45 rows=156873 width=131) (actual time=13578.991..15239.979 rows=3163907 loops=1)
               Sort Key: t.hash, t.name
               Sort Method: external merge  Disk: 465696kB
               ->  Nested Loop  (cost=1.13..5292425.73 rows=156873 width=131) (actual time=0.080..2314.057 rows=3163907 loops=1)
                     ->  Index Scan using idx_job_finished on job j  (cost=0.43..443.73 rows=449 width=16) (actual time=0.062..9.326 rows=12109 loops=1)
                           Index Cond: ((repository_id = $0) AND (status = 3) AND (date_finished > (now() - '14 days'::interval)))
                     ->  Index Scan using unq_testcase_hash on testcase t  (cost=0.70..11668.60 rows=11755 width=147) (actual time=0.008..0.129 rows=261 loops=12109)
                           Index Cond: (job_id = j.id)
 Planning time: 0.462 ms
 Execution time: 15990.365 ms
(18 rows)

bors r+

getsentry / zeus

[Merged by Bors] - ref: Switch repo tests query to be a (somewhat) faster join #391