Rewrite KDE visualization using new dclab implementation

[paulmueller]

Depends on: DC-analysis/dclab#153
Original issue: #110

[chrherold]

Maybe some comments for discussion here:

primary goal: base the visualization of the density of events in a region on a useful histogram map that is used for coloring points as well as drawing contour lines

(means discontinue KDE estimates for Shape-Out2, as they are often misleading if setting are wrong or population is not described by a gaussian distribution)

some considerations

• find a good and automated way to set the correct spacing of the grid describing the density
  • deal with outliers
  • one grid per plot panel? (!)

secondary goal: make plotting faster

Currently, by built in design, KDE is calculated using all points in a plot (downsampling makes it faster)
For determining the grid, a downsampled subset (5k to 10k?) could be used even if all points are plotted.
This should speed up plotting of datasets with many events without losing significant information.