Kliff

- add support for the geodesic Levenberg-Marquardt minimization algorithm
- add command line tool `model` to inquire available parameters of KIM model

- add RMSE and Fisher information analyzers
- allow configuration weight for ML models
- add write optimizer state dictionary for ML models
- combine functions `generate_training_fingerprints()` and
`generate_test_fingerprints()` of descriptor to `generate_fingerprints()`
(supporting passing mean and stdev file)
- rewrite symmetry descriptors to share with KIM driver

- MPI parallelization for physics-based models
- reorganize machine learning related files
- various bug fixes
- API changes \* class `DataSet` renamed to `Dataset` \* class `Calculator` moved to module `calculators` from module `calculator`

- KLIFF available from PyPI now. Using `$pip install kliff` to install.
- Use SW model from the KIM website in tutorial.
- Format code with `black`.

First official release, but API is not guaranteed to be stable.

- Add more docs to {ref}`reference`.

Pre-release.