Neural collapse vs. low-rank bias: Is deep neural collapse really optimal?