Implement log-scale KDE visualization

[paulmueller]

The kernel density estimate (KDE) is computed on a linear scale, which looks unintuitive on a log-scale. Log-scale KDE has been implemented in dclab (DC-analysis/dclab#55).

• In order to implement it in Shape-Out, a new interpretation of / solution for KDE accuracies in the case of log-scale is needed. EDIT: Because of this, log-scale contours will only be available in expert mode for now. EDIT EDIT: kde/contour accuracies are now reset when switching from log to linear or back.

This affects

• KDE of contour plots
• KDE of scatter plots
• KDE used when downsampling data for visualization purposes
• KDE used when downsampling data to specific size ??? (randomly downsampled)

[paulmueller]

This does not seem to be possible with chaco at the moment.

EDIT:

• Log-scale KDE for scatter plots is now available using the dclab implementation (mentioned as a bugfix in changelog).
• Log-scale contours are a different story: The contour plots from chaco were not usable (small offset on linear scale, huge offset on log-scale, see f573815), so we are now computing contours in dclab>=0.10.1 and displaying them correctly with chaco as simple line plots

[paulmueller]

An new interpretation of kde/contour accuracies is circumvented by simply resetting the contour accuracy when switching from/to log/linear (7de3bc0).