Config file subsections -- init and fit

[ceholden]

To make our configuration files better pair with the scikit-learn estimator API, each timeseries model must now be configured using an init and fit subsection.

Example from some made up estimator:

MadeUpModel:
    init:
        threshold: 3
        likelihood: binomial
        sided: two-sided
    fit:
        sample_weights: '1 / vza'

This greatly improves the flexibility of the configuration files because it explicitly declared what configuration settings are for what methods. This can be extended for other methods (e.g., transform) for scikit-learn-like APIs.