Add simulation based calibration tooling

Goal

The goal here is to add simulation-based calibration model checking to the forecasttools toolbox. This then provides a set of checks for the self-consistency/correctness of a Bayesian model + a posterior sampling method. This is aimed as acting as a unit test for models.

Proposed solution

My proposal is to smooth any issue with using SBC to work with tidybayes (this is likely already available) objects that are structured over multiple sbc trials. In reality, most models we will try outputing an Arviz object so the idea will be to leverage Arviz to tidybayes conversions we already have in place.

At least as a first pass, the user will be responsible for generating the sbc trials.

As a diagram the expected eventual workflow is as follows:

graph TD
    Generate_data[Generate datasets from model] -->|inference| posterior_draws[Posterior draws for each dataset]
    posterior_draws -->|Arviz| make_data_set[Make InferenceData object over datasets]
    make_data_set -->|Serialise| serialise[Serialize to netCDF]
    make_data_set -->|direct convert| convert[Convert to tidybayes]
    serialise -->|deserialise and convert| convert
    convert -->|SBC| sbc[Run SBC checks and return report]

Start by locating the existing Arviz-to-tidybayes conversion code in forecasttools and reviewing how tidybayes objects are structured across SBC trials. Compare that path with the SBC workflow described in the issue; done would mean accepting user-generated trials and returning simulation-based calibration checks or a report, but no file or test paths are identified.