When creating a CSV in Excel, it by default includes (what I think is) a BOM marker at the top of the file, to ensure that it’s parsed as UTF-8. Unfortunately Discourse doesn’t handle this (and doesn’t read the file as UTF-8!) so the first email address always fails with an error like this:
(Note that the “j” is the first letter of the actual email address.)
I think Discourse should strip/ignore the BOM on CSVs – it’s probably pretty common that people would put these together in Excel, after all.