On the rates of asymptotic normality for recursive kernel density estimators under ϕ-mixing assumptions