Clarify the difference between the `FeatureArray` wrapper and the wrapped types

[aazuspan]

This came up in discussion with @grovduck in #103.

The FeatureArray class is a wrapper around one of several types (currently np.ndarray, xr.DataArray, xr.Dataset, and pd.DataFrame) that provides a consistent interface and tracks NoData for use in estimator methods and other ufuncs. However, the documentation and API frequently refer to the wrapped types as feature arrays, e.g. from the Usage Guide:

sklearn-raster supports applying estimator methods to rasters stored in a variety of formats, referred to collectively as feature arrays

And from the API:

• FeatureArrayType is a typevar around the wrapped types.
• FeatureArray.from_feature_array constructs a wrapper from a wrapped instance.
• FeatureArrayEstimator methods work with the wrapped objects, not `FeatureArray.

Now that ufuncs (and by extension FeatureArray) are becoming a public API rather than an internal implementation within FeatureArrayEstimator, it would be valuable to decide exactly what a "feature array" is (the wrapped object or the wrapper?) and make sure the naming and documentation are consistent.