The impact of random‐effect misspecification on percentile estimation for longitudinal growth data