Allow GP data to be held in a Pandas Data Frame

[edaub]

R users typically provide their data to a regression model in a Data Frame. Similarly, most statistical Python packages use data frames when building regression models. This has the advantage of avoiding need of an input dictionary when parsing from a formula.

This should probably be the preferred mechanism for feeding data into the GP class, though we may still want to support inputs/targets as before.

One challenge: multiple outputs from a single dataframe is probably more difficult. We would certainly need a way of pulling out all of the targets so they aren't used as the inputs. Mean functions would need to be a list of string formulae identifying which targets are to be fit. There are probably a few other things to worry about on this front.

I'm currently building a prototype to test several specific changes at once to be sure they work nicely in concert. See the associated project board, which collates these issues together.