Deep neural operators can serve as accurate surrogates for shape optimization: A case study for airfoils