batchCop writes data to TiDB just using one thread.

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

    {
        if (!dag_context.retry_regions.empty())
        {
            coprocessor::BatchResponse response;
            for (const auto & region : dag_context.retry_regions)
            {
                auto * retry_region = response.add_retry_regions();
                retry_region->set_id(region.region_id);
                retry_region->mutable_region_epoch()->set_conf_ver(region.region_conf_version);
                retry_region->mutable_region_epoch()->set_version(region.region_version);
            }
            writer->Write(response);
        }

        auto streaming_writer = std::make_shared<StreamWriter>(writer);
        TiDB::TiDBCollators collators;

        std::unique_ptr<DAGResponseWriter> response_writer = std::make_unique<StreamingDAGResponseWriter<StreamWriterPtr>>(
            streaming_writer,
            std::vector<Int64>(),
            collators,
            tipb::ExchangeType::PassThrough,
            context.getSettingsRef().dag_records_per_chunk,
            context.getSettingsRef().batch_send_min_limit,
            true,
            dag_context);
        dag_output_stream = std::make_shared<DAGBlockOutputStream>(streams.in->getHeader(), std::move(response_writer));
        copyData(*streams.in, *dag_output_stream);
    }

2. What did you expect to see? (Required)

3. What did you see instead (Required)

batchCop send blocks just using one sync grpc thread.
the thread should encode then send to tidb/tiflash.
TODO:
move the streamWriter in the stream.in.

4. What is your TiFlash version? (Required)

master