Add non-parametric reference date model

[seabbs]

Currently only the parametric reference date model has been implemented. For completeness it would be useful to update the model to include the non-parametric component defined in the model definition vignette.

This requires:

• Adding an effects argument and data list to the input
• Adding new initialisation parameters
• Adding another layer of logit hazards in the stan code
• Checking the quick start and adding commentary on this approach to the case study.
• Make it possible to only have a non-parametric model
• Consider what the default model should be?

[seabbs]

A simple way to do this is to only allow non-parametric hazards to be set across all reference dates. This would mean having a design matrix with rows equal to the maximum allowed delay and arbitary feature. This vector of delays could then be exposed to the user to enable them to build models. The features exposed would need to be those that do not vary with time.

This formulation of the model, whilst limited, is comparable to what others have implemented and use. It may be possible to extend it in the future with additional complexity.

[seabbs]

I've planned out how to implement this feature in branch seabbs/issue4 with a view to implementing a prototype in the next few months.

[seabbs]

Will be addressing this in the next week if anyone is interested in reviewing that would be great.

Once this is in place it will also open up the question as to what the new default model should. I think that discussion should probably be had in a issue but I think I would likely favour a non-parametric default model - potentially a weekly random walk.