Simulation-based testing

[afmagee42]

⚠️ Work in progress ⚠️

This PR adds simulation-based tests of performance as package data to provide users context on how well inference works in some parameter regimes of interest (and one or two of perhaps less interest, with larger $R$ and $k$). Inference summaries are stored in the form of point (posterior median, maximum likelihood) and (95% credible and confidence) interval estimates.

An in-development vignette uses this data to provides a sense of coverage and bias in practice when using the model. (It doesn't cover cases where the model is misspecified, but that universe is far larger and out of scope.)

To do all this, this PR adds a make-based workflow to data-raw which can be run to simulate across a grid of $R$, $k$, and dataset size, as well as analyze with both Bayesian inference and maximum likelihood (both under default settings). The use of make and bash scripts which exist only to call R may not be strictly necessary. However, previous attempts at a purely R-based workflow resulted in a variety of failures, while this appears to be working.