Survival models

I'm opening this issue so there is a forum in which to discuss progress on getting survival models into rstanarm.

I've started a new survival branch here. It currently has a stan_surv modelling function, calling a surv.stan model file. The modelling function returns an object of class stansurv that inherits stanreg.

The current task list looks something like:

• print method for stansurv
• summary method for stansurv
• prior_summary method for stansurv
• basehaz_summary method for stansurv
• posterior_survfit method for stansurv
• log_lik, loo, and waic method for stansurv
• plot method that includes plot type for baseline hazard
• add option for rcs splines for the log hazard (i.e. instead of B-splines)
• get the piecewise constant baseline hazard from stan_jm working with stan_surv
• attempt to implement a constrained log-likelihood by just shifting hazard (setting equal to 1E-8 or similar) when the hazard strays negative, see section 2.2 of this paper
• add functionality for time-dependent effects (i.e. non-proportional hazards) -- should this use survival package's tt approach, or something else (e.g. a tde() special function for use in the formula)
• add AFT models
• add semi-parametric model (possibly like this paper)
• add frailty models

Start with the survival branch and read the current stan_surv function, surv.stan model file, and stansurv methods. Choose one unchecked task, such as basehaz_summary or rcs splines, then confirm its expected behavior with the existing completed methods and tests; done means that task works for stansurv without broadening the scope.