OK, after some tests:
The trick is in surrounding text in your csv with quotes.  Like that spark can actually make a difference between a missing value and an empty value.  Missing values are null and can only be converted to something else implicitely (by using coalesce f.e.).
When a column contains '', nullvalue = "''" will create an empty value and not null.
The same for emptyValue if you want.
Not sure if it is workable for you though.