Earlier quoted context omitted.
The record separator isn't on people's keyboards, so it's less likely to show up where it's not expected. Also it's less likely to legitimately occur in something like a name, so there are many users of CSVs who can say they will never need to consider data containing a record separator, and they will be right more often than those who never consider data containing a comma. Of course, the fact that record separators…
I can’t think of a case where someone would write a control character like that into something intended for text on purpose. So you might as well disallow it.
It is still valid to disallow the ASCII control characters, one just has to make sure that it is done comprehensively, in all places users may input them. But that's not created by using ASCII control characters, that's a consequence of the "ban the control characters entirely" approach regardless of what the control characters are.
It's neat when you can get away with it, but I generally prefer to define a robust encoding scheme instead. A minimal one like "replace backslash with double-backslash, replace control characters with backslashed characters" and "replace backslash sequences with their control characters, including backslash-backslash as a single backslash" can be inserted almost anywhere in just a few lines of string replace (or stream processing if you need the speed). The only tricky bit is you need to make sure you get the order correct or you corrupt data, and while I've done this enough to have it almost memorized now I do recall feeling like the correct order is backwards from what I naturally wanted the first few times. But it is simple and robust if you get it right.