Initialisation model and improved unit testing for distributions

[SamuelBrand1]

Main change

This PR makes a simple initialize_incidence Turing model constructor available, but doesn't implement using it. This is because we are likely to refactor the API from this point.

Minor change

Unit tests on distributional usage were relying on direct analytic solutions; to speed this up I've added HypothesisTests as a dependency to the test environment (not the package environment).

Usage

• For expected continuous distributions, random variables sampled from EpiAware functions are compared against the expected distribution using a Kolmogorov-Smirnov test HypothesisTests.ExactOneSampleKSTest e.g here.
• For expected discrete distributions, random variables sampled from EpiAware functions are compared against the expected distribution by also generating samples from the expected distribution and comparing two sample averages using one way ANOVA HypothesisTests.OneWayANOVATest, and variance using a F test HypothesisTests.VarianceFTest. Implementation example here.

For hypothesis tests the test fails at p < 1e-6, e.g. if there is a correctly implemented test the failure chance is one-in-a-million. Typically, this means that sampling from the model should be in order of a few thousand to avoid potential false passes.

[seabbs]

This looks good to me. Really nice.