Stats: Allow partial pooling of sample-specific prevalence estimates

[dp-rice]

In the current model, each sample has its own "true prevalence" parameter and they are independent of one another. In the previous version, all samples shared a single "true prevalence" parameter.
Ultimately, we would like to have a model that allows for both shared and independent variation, e.g.:

$\theta_i \sim Normal(\theta, \tau)$

$\theta \sim Normal(\mu, \sigma)$

where $\theta_i$ is the log true prevalence in one sample and $theta$ is a global average prevalence. We could extend $\tau$ to be a covariance matrix.

In this setting, our prevalence data would be emitted directly from either $\theta_i$ or $\theta$ depending on whether its local to a sample or global.

Ultimately, we might want the thetas to be a Gaussian process over time and locations and have both sequence data and prevalence data be emitted from this process.

[dp-rice]

Closing as not worth the effort.