I don’t see what does it change.
Now what kind of a user expectation it is? Do you even think this can be considered a “user expectation”?
Try to imagine, how many of the hundreds of millions who encounter CSVs, when approached and asked “what do you expect when you open, edit and save a CSV”, would give such a reply? I think, one - you. Everyone else would be like “I expect that the result would be accepted where I send it, and the data gets correct from it, and my main goal of providing them the necessary information is fulfilled”. People expect fitness for a purpose, for a given workflow, and that always includes some third parties that define some “subformat” of CSV, which is implicitly defined by the initial content of the CSV received by the user. Making the program to handle that correctly, is millions of times more important than generating some abstract “vanilla csv” (it one even imagines there’s such a thing - the CSV RFC is not a kind if standard, it has informational status - compare e.g. to RFC 1738, with is “Standards Track”).