On a Skew t–discrete Gamma Alternative to the Normal-Poisson Model for Clustered Count Outcomes