function to calculate Gelman Bayesian R2

[sebdalgarno]

I think can calculate from fit in new_expr, i.e.,

var_y_pred = variance(y_pred);
  var_residual = variance(to_vector(count) - y_pred);
  R_squared = var_y_pred / (var_y_pred + var_residual);

think about whether this is what R2 in embr is already doing