I love performance tunings. I'm drawn here by your benchmark results. By the way, I think it would much better if you describe the reason why this library is faster than `pg`. Because I'm personally curious, and if do not explain why, there may be some people who doubt it.