On the maximum deviation between the histogram and the underlying density