Sql. Improve conversion to BinaryRow in UpdatableTableImpl

As for now, UpdatableTableImpl uses RowAssembler to re-assemble row prior to send it to the table. The problem here is this is not very efficient because of number of reasons. First, the most simple constructor of RowAssembler is used, this causes tuple builder to preallocate 4kb of buffer, which sometimes is too much. Second, usage of RowAssembler implies double (de-)serialization since every field should be read from source row and write back to target tuple.

Given that sql row now natively supports to binary tuple conversion, depending on the schema of a table we may reuse that binary tuple or reshuffle fields avoiding deserialisation by reading raw bytes.