Implement new `Corrector` API design

During a brainstorm today (by @christinahedges, @nksaunders, @barentsen), we provisionally decided to deprecate LightCurve.correct() in favor of an object-oriented approach centered on an abstract Corrector class which has three key methods:

Corrector(**data_required):
    .correct(**options) -> LightCurve
    .diagnose(**options) -> Figure
    .interact() -> Bokeh widget

Specific implementations of this class would include SFFCorrector, PLDCorrector, and PCACorrector. The rationale behind removing lc.correct() is three-fold:

1. Different correction methods require different input data and different options, which is difficult to expose as a single method.
2. Forcing pixel-level decomposition (PLD) methods to operate on a LightCurve objects is awkward because the method requires pixel-level data.
3. By exposing Corrector class objects, we will encourage users to research the correction methods being applied and, if necessary, develop their own Corrector subclass.

In summary, the old syntax:

lc_corrected = lc.correct(method="sff", **options)

…will be replaced by the new syntax:

lc_corrected = SFFCorrector(lc, centroids).correct(**options)

[barentsen]

I’d like to leave this open until we have added a diagnose() method to both SFFCorrector and PLDCorrector. (Even if all it does for now is return an informative summary plot.)

[barentsen]

Closing this issue due to inactivity. Improvements on the Corrector classes continue to be made, e.g. over in #525.