Clusterking

Fixed

- Critical issue with chi2 metric: Correlations were not properly handled and number
of degrees of freedem were not correct (should be number of bins - 1, not number
of bins).

Removed

- `decorrelate` argument to `data` method of `DataWithErrors` class, because it
was incorrectly implemented and probably does not have reasonable use cases

Deprecated

- python 3.5 will not be supported for future releases

Added

- Find sample points or benchmark points closest to a given point in parameter space
- Matrix plot of distances of benchmark kinematic distributions

Added

- Configuration of plot labels
- Many improvements and added features for stability tests

Added

- Stability tests

Added

- Scanner: Sample distribution (rather than binning it)
- Configure variable labels on axes

Changed

- Follow command design pattern and change interfaces accordingly:
More formal pattern of worker and result classes
- Make physics package its own package
- Make matplotlib dependency optional
- X axis of distribution plots now shows the kinematic variable, rather than the bin number

Fixed

- Fix calculation of poisson errors
- Compatibility with different matplotlib and pandas versions

Changed

- The two output files ``*_data.csv`` and ``*_metadata.json`` have been replaced by **one** output file in SQL format. This also simplifies loading and saving Data objects and furthermore reduces the required disk space.