need better support for cross validation / typical fitting workflow

[bcdaniels]

The original Sir Isaac code performs model selection using a modified Bayesian information criterion. This makes sense in cases with limited data, but in many applications, we have enough data that it makes more sense to use cross validation (fit to a subset of the data and select a model based on its out-of-sample performance).

In particular, this is what we did in applying Sir Isaac to C. elegans behavior data. (Note to self: This was implemented in gatherResults.goodnessOfFit_model.) We should make this easier to do within Sir Isaac.

This could also be an excuse to automate what has become the usual model fitting workflow, in which we first fit to a very small subset of data and then gradually add data, refitting at each step.