Kafe2

* Performance optimizations. In particular the inversion of the covariance matrix has been replaced with a cholesky decomposition and scipy.linalg.solve_triangular .
* Fixed a bug where sometimes old parameter values were being used. Fit results changed slightly.
* Fixed various minor bugs.

* Option to set logarithmic scales for plots.
* Chi2 cost function now adds the logarithm of the determinant of the covariance matrix to its cost to reduce bias. Can be used by custom cost function too by setting add_determinant_cost=True in the constructor.
* Some gof/ndf display logic fixes.

On December 7th iminuit 2.0 was released.
kafe2 does currently not support the new interface. kafe2 won't use iminuit if a version >2 is detected

* Skip using iminuit if iminuit>2 (until we support the new interface)

If you still want to use iminuit with kafe2, please install iminuit 1.x with `pip install 'iminuit<2'`

This "2.2.2-1" release fixes an issue from "2.2.2" when retrieving the version from iminuit, when iminuit is <2.

* Added support for iminuit 2.
* Fixed parameter constraints not increasing ndf.
* Fixed shared errors for multifits.

Fixed several bugs, mostly related to MultiFit:

* Fixed a bug where some kafe2-internal caches weren't properly cleared.
* A debug log message being printed as a user warning when adding shared errors to a MultiFit.
* Issues when adding more than one shared error.
* Issues when creating a MultiFit from different fit types.
* Shared errors sometimes being added to the wrong fit.
* Shared x errors sometimes using the wrong scale for calculating derivatives.

Relative to the last pre-release:

* Expanded documentation.
* Fixed documentation LaTeX build not working.