feature/car

Starting an issue to track the feature/car branch which introduces the conditional autoregressive (CAR) class of spatial models into rstanarm.

Currently it looks like,

            ┌─── stan_besag.R ────┐
user ───────┤─── stan_bym.R  ─────├──── stan_spatial.fit.R ──── spatial.stan
            └─── stan_bym2.R ─────┘

• stan_besag.R is the intrinsic conditional autoregressive (ICAR) model (i.e. just a GMRF with a sum-to-zero constraint.)
• stan_bym.R is a CAR model which involves a convolution between the GMRF and noise.

Stuff to implement for the models above,

• Stan model
• QR Decomposition
• Centering covariates
• Optimize/VB
• Print/Summary
• Methods
• Likelihoods (gauss/binom/pois/neg_binom/gamma)
• Priors
• Prior summary
• Stanreg
• Posterior predict and pp_check
• Loo
• Documentation
• Vignette(s)
• Tests
• All sparse matrix operations should use the Matrix package
• Drop INLA dependency (currently required to compute scaling_factor in stan_bym with INLA::inla.qinv)