ClickHouse 3970: Add setting min_merge_bytes_to_use_direct_io

[alesapin]

Hack in code :(. But changes are very small. The only good solution is to split MergeTreeBlockInputStream, as said in comment https://github.com/yandex/ClickHouse/blob/master/dbms/src/Storages/MergeTree/MergeTreeBlockInputStream.h#L15.

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

[alexey-milovidov]

Ok.

Hack in code :(. But changes are very small. The only good solution is to split MergeTreeBlockInputStream, as said in comment https://github.com/yandex/ClickHouse/blob/master/dbms/src/Storages/MergeTree/MergeTreeBlockInputStream.h#L15.

Note that the methods createReadBufferFromFileBase and createWriteBufferFromFileBase accept both estimated_size and aio_threshold as parameters, with the only purpose to do simple comparison.

I think, it will be better to pass just single parameter named use_direct_io, and move all logic with thresholds outside.