nir/lower_io_to_vector: allow FS outputs to be vectorized