Gravar-mail: Deviation from the matching law reflects an optimal strategy involving learning over multiple timescales