Write class for KDE estimation

[paulmueller]

The current implementation of KDE computation (partly in rtdc_dataset.core) is not very clean. It would be best to have a new RTDCKDEEstimator class that can:

• be instantiated with parameters such as kde_method="histogram"
• uses the same KDE for scatter plot data as for contour data
• this implies that for scatter plot data, the KDE is computed on a grid as well (and not for each point)
• it would be nice to see timing information that compares a large and a small dataset with the methods implemented currently
• there should be a method that refines the class parameters until the contour is smooth (no kinks, probably only applies to the histogram method)

This will be a breaking feature, so it would probably make sense to deprecate it properly in rtdc_dataset.core and dclab.kde_methods.

Related issues:

ZELLMECHANIK-DRESDEN/ShapeOut2#110

[paulmueller]

Consider comments in ZELLMECHANIK-DRESDEN/ShapeOut2#111