pubsubsink waits for the last message to be completely delivered before sending the next one.
This is inefficient, because if we don't batch the requests, the throughput is limited to 1/write_latency .
A solution would be to .Get the result only once all messages have been published.