Pylandstats

- fix sum(level) -> groupby(level).sum (pandas deprecation)
- dropped six for metaclasses (using direct ABC inheritance)

- GitHub Actions tests in develop branch too

- added methods to compute information theory-based metrics by Nowosad et Stepinski (2019).
- using pre-commit with black, isort, flake8 and pydocstyle
- CI (tests) and CD (release to PyPI) with github actions

- improved exception caching in compute metric data frames methods
- init `BufferAnalysis` with `Landscape` with non-None `transform`
- test required init args in `Landscape` and `ZonalGridAnalysis`
- test `ValueError` in data frames with patch/class/landscape-only metrics
- several corrections in docstrings/comments/error msgs
- fix: removed `SpatioTemporalBufferAnalysis._landscape_metrics_df` (dropped this kind of caching in v2.0.0)
- using python3 `super` with no arguments
- added `neighborhood_rule` argument (to choose 4/8-cell adjacencies)
- fixed legend in `plot_landscape` and added `legend_kws` arg
- fixed multilandscape plot with one landscape, e.g., 1-zone analysis
- fixed values of zone_arr (to id zones) to avoid nodata=0 confusion
- default nodata (when providing numpy arrays) from settings module

- `ZonalAnalysis.attribute_name` from `masks` when providing geoseries
- avoid reprojecting equivalent CRS (pyproj `__eq__` vs. `is_exact_same`) in `ZonalAnalysis`

- fill `NaN` values according to the metric in `MultiLandscape.compute_class_metrics_df`
- consistent imports so that there are no direct class/function calls, and no module abbreviations in the documentation
- zonal analysis with more general `masks` argument, which also accepts vector geometries (and replaces the now deprecated `masks_arr`)