Chaining `InferenceProcess` together

build away for the output of one InferenceProcess to inform the beginning of another.

two examples of where this may be used.

1. chaining inference together into successive epochs (fitting one year of flu at a time for example)
2. initialize an MCMC process using the learned parameters from a faster SVI fit.

Both of these problems require one process to flow into the next, in slightly different ways. 1. requires the loading of a previous fit's final compartments into the initial state of the next epoch. 2 requires moving the mass matrix from SVI into MCMC to speed up fitting.

This feels like two similar but slightly different problems that may be solved in a single interface, but this ticket may need to be split into two if we find the problems to be different enough from one another.

Start by locating the InferenceProcess entry point and reading how an inference run exposes its final compartments, learned parameters, and mass matrix. Compare the successive-epoch and SVI-to-MCMC flows to determine whether one interface can support both or whether the issue should be split. Done means one process can provide the required state or mass matrix to the next process.