Replace the underlying eigenvector write with a version that can write eigenvectors with indices in the range [ki, kf) as with the eigenvector read functions
Add overloads that reproduce the full-pack writes, similar to the full-pack read overloads, so that existing workflows do not break