With '%', non-ascii worked because the '%' automatically got promoted to
unicode.  With format that doesn't happen, which led to encoding errors.  This
fix goes back to using %, and adds a test to make sure non-ascii string values
work in iterdump.