Model specific priors capability

[SamuelBrand1]

We have been concentrating on the infrastructure of the pipeline, but I think the priors need some reasoning upon because for different scenarios they end up having different interpretations.

For example, the typical variation over time for the exponential growth rate $r_t$ is going to be on a different scale to $\log(I_t)$ the log-incidence. Our prior beliefs should reflect this.

To be able to implement this we probably to vary away from a default constructor for priors and towards a constructor that dispatches on the IGP being selected in the list of possible model combinations.

EDIT: Adding a smaller task list to get here

• Plotting methods for prior predictive. #455
• Full prior predictive plotting step in pipeline #468
• Change saved prior predictive data #477
• Analysis of what is causing models to break. #481
• Implement model specific priors #522

[SamuelBrand1]

A particular problem here is that we also have a (current default) 35 days look back for inference... this means that at least some of the priors should depend on state at beginning of lookback window (Or we revert to just fitting on whole time series).

[SamuelBrand1]

This is part of resolving #4 .