use case for threading and sparsity

[paciorek]

I think that Laplace is a good use case to experiment with the new threading and sparsity capability. This is just a note with some details for down the road.

The Bernoulli GLMM example in Stringer et al. Sec 5 (with comments on parallelization in Table 1) is replicated (but with simulated data) for use with Laplace/nested approximation in my nested-bern.R which will be committed to tests/testthat in nimbleQuad (or see code in NCT issue 577).

Also note that in this GH repo for the Stringer et al. paper code there is code for using INLA or glmmTMB (if one sets doFit=TRUE) on this dataset. Stringer notes that both INLA and TMB/glmmTMB can use multiple threads to evaluate the log-lik in parallel (see the parallel argument in ?glmmTMBControl).

So the idea would be to try to modify our Laplace to make use of threading and compare to glmmTMB and INLA (and possible aghq package).

Regarding sparsity, INLA and (seemingly) glmmTMB can make use of sparsity, so similar thought about seeing what we can do on that front in Laplace as well.