Personally I think the benefit of sharing fieldsep is not worth these problems, but I'm waiting for the discussion to continue with more opinions.
Apologies in advance if I mis-represent someone's position.
It seems like having a dedicated option is the consensus opinion. Daniel (the original author) and I both prefer it, Pavel will accept it. Fabien is opposed.
Peter E. was opposed, wanting to leverage both fieldsep and recordsep, but hasn't chimed in recently. His opinion at this point might push this over the edge since he is also a committer.
I would probably suggest maybe just calling it "\pset separator" to match the "S" in "cSv" and not have any name overlap with the fieldsep variable used with unaligned mode. The user will have to learn anything and being more distinct should help the process. We would not consult recordsep though the end-of-line choice should be platform dependent.
-1. The difference between fieldsep and separator is not clear, and the relation between separator and csv is not clean too.
fieldsep_csv or csv_fieldsep is not nice, but it is clear.