Observation process as a Turing model

[SamuelBrand1]

At the moment the observation process is embedded inside the make_epi_inference_model function/constructor

Rt-without-renewal/EpiAware/src/models.jl

Lines 21 to 27 in 22a5f5d

	#Predictive distribution
	case_pred_dists =
	(epimodel.data.delay_kernel * I_t) .+ pos_shift .|>
	μ -> mean_cc_neg_bin(μ, neg_bin_cluster_factor)
	#Likelihood
	y_t ~ arraydist(case_pred_dists)

The reason for this is that this project envisions using only one type of observation model: a time-invariant delay distributions between infection and reference time of a case.

The problems are that:

• This doesn't promote modularity/extensibility, because it hard fixes an modelling approach.
• It lacks flexibility within the short term focus of this project. For example, we might want to generate some time series of infections before case observation; this being a standard initialisation approach to inference problems on case time series (see also Initialisation infection generation processes #40 ).

Proposed solutions

Abstracting the observation process to a Turing model with options.

[SamuelBrand1]

This was closed by #44