Gaussian Quadrature and Laplace Approximation

[ParadaCarleton]

With large datasets, MCMC can get very slow. Luckily, approximations based on numerical integration usually get better as the sample size increases, and can work in some situations where MCMC is impractical. Luckily, the code for this is already implemented in MixedModels.jl and just needs to have priors added for it to work; we should probably take a look at it.

[storopoli]

Do we want to implement this? Users could just MixedModels.jl and leave TuringGLM.jl to being exclusively a "formula interface to Turing.jl"?