It would be great if a comparison to the rust csv library or xsv [1] could be added - it's single threaded, but it has a zero copy option as far as I know. I've used it to parse / filter large PostgreSQL log files (20GB+) with great success.
The other bit to mention is `xsv index`. If you can afford to make one single threaded pass over the CSV data, then future operations truly do become embarrassingly parallel. It's a simple low-tech solution, although the UX is worse because you need to take a manual step.
As for zero copy, could you say where you saw that? I don't think it makes sense to describe xsv that way. The csv parser does have a lower level zero allocation API, but there is no zero copy API.