Do we need decay-preserving error estimate for solving parabolic equations with initial singularity?