Gravar-mail: Heavy-tailed prior distributions for sequence count data: removing the noise and preserving large differences