This might go too far by using `ActiveRecord::AttributeSet::Builder`
but unfortunately that's how Rails is doing that to unify the results
from different database adapters. I don't know why it's built into
the model so tightly while it could just stay as in connection.

This is a class where we could use in the migration,
creating a database connection pool with arbitrary
pool size. In general, the only use case would be:

``` ruby
Gitlab::Database::ConnectionPool.with_pool(2) do |pool|
  pool.execute_async('DELETE FROM table0;')
  pool.execute_async('DELETE FROM table1;')
  pool.execute_async('DELETE FROM table2;')
end
```

This would run the queries with ActiveRecord's `with_connection` and
`connection.execute` in threads, and wait until all the queries are
done, and finally gracefully close the connection pool.

Rename Files::DeleteService to Files::DestroyService

See merge request !9110

Resolve "Improve blockquote formatting in notification emails"

Closes #28029

See merge request !9157

Convert some vendored javascript assets to npm modules

See merge request !8990