key16 aggregation method should be two level

[alexey-milovidov]

Currently, key16 aggregate states cannot be merged in parallel:

milovidov-pc :) SELECT number % 10000 AS k, uniq(number) AS u FROM numbers_mt(1e9) GROUP BY k ORDER BY u DESC LIMIT 10

SELECT
    number % 10000 AS k,
    uniq(number) AS u
FROM numbers_mt(1000000000.)
GROUP BY k
ORDER BY u DESC
LIMIT 10

Query id: d3e22e17-1a32-4615-bf1a-a2da6e0510eb

    ┌────k─┬──────u─┐
 1. │ 4759 │ 101196 │
 2. │ 4587 │ 101079 │
 3. │ 6178 │ 101034 │
 4. │ 6567 │ 101032 │
 5. │ 9463 │ 101013 │
 6. │  298 │ 101009 │
 7. │ 2049 │ 100993 │
 8. │ 8167 │ 100989 │
 9. │ 5530 │ 100973 │
10. │ 1968 │ 100973 │
    └──────┴────────┘

10 rows in set. Elapsed: 62.793 sec. Processed 1.00 billion rows, 8.00 GB (15.93 million rows/s., 127.40 MB/s.)
Peak memory usage: 11.30 GiB.

milovidov-pc :) SELECT 0 + number % 10000 AS k, uniq(number) AS u FROM numbers_mt(1e9) GROUP BY k ORDER BY u DESC LIMIT 10

SELECT
    0 + (number % 10000) AS k,
    uniq(number) AS u
FROM numbers_mt(1000000000.)
GROUP BY k
ORDER BY u DESC
LIMIT 10

Query id: e6a24292-54cf-47cb-8e39-81584736d41a

    ┌────k─┬──────u─┐
 1. │ 4759 │ 101196 │
 2. │ 4587 │ 101079 │
 3. │ 6178 │ 101034 │
 4. │ 6567 │ 101032 │
 5. │ 9463 │ 101013 │
 6. │  298 │ 101009 │
 7. │ 2049 │ 100993 │
 8. │ 8167 │ 100989 │
 9. │ 5530 │ 100973 │
10. │ 1968 │ 100973 │
    └──────┴────────┘

10 rows in set. Elapsed: 8.547 sec. Processed 1.00 billion rows, 8.00 GB (116.99 million rows/s., 935.95 MB/s.)
Peak memory usage: 10.09 GiB.