Make model prediction-ready

[AlexAndorra]

My next steps to make the model appropriate for predictions are:

• add popularity and econ variables as predictors (i.e, fundamentals)
• add pollsters' historial biases
• add the number of undecideds to the model
• add the hierarchical structure for past elections
• [ ] develop MvZeroSumNormal to estimate the covariance between parties (for now it works -- see commented lines in the model -- but is sub-efficient sampling-wise because the centered parametrization would work better)