First official release

* Disable hyperpriors default setting so that the estimated priors

  on the means just use the maximum likelihood estimates.
* Added four more iterations on the Gauss-Siedel iterations of the
  template updates.
* Changed the tolerance for assessing convergence of the overall
  algorithm.