Exploiting hypersparsity in the revised simplex method