On consistency of kernel density estimators for randomly censored data: rates holding uniformly over adaptive intervals