What to do with Model.fit_reference_beta()

[jgallowa07]

We have a Model method for fitting the beta parameters (warmup) to the reference data. Currently this just solves the linear problem A x = b, but in reality, if it's to be useful for the rest of the model, it should optimize the linear model specifed (with $\beta_0$ and potentially $\alpha_d$). Alternatively, if we never think this will be useful, it could be deprecated.