Pyextremes

- fixed an error caused by passing `pandas.Series` with duplicate datetime indices to `pyextremes.EVA` and calling `.get_extremes`

- Fixed bug in the `plot_corner` function which previously returned incorrect list of axes.
- Several minor code optimizations.

- Added `test_ks` method to `EVA`. Allows to run the Kolmogorov-Smirnov test on the fitted model.
- Updated the way declustering in the `get_extremes_peaks_over_threshold` function happens. Clusters are now identified independently. Previous version could result in multiple extreme events within the same cluster if the cluster had multiple peaks separated by distances larger than `r`. This behavior was incorrect.
- Fixed issue when `get_return_periods` function could produce output linked to original extremes, thus allowing implicit mutation of original data.
- Updated old and created new plotting functions for threshold selection in `pyextremes.tuning.threshold_selection`.
- Minor backwards-compatible tweaks and optimizations.

v.2.2.2
Hotfix for new `scipy` version 1.7.0 - fixes issue when getting confidence bounds while using the MLE model. Some distributions provided by scipy no longer return a fit method, but instead a reference to a wrapper which resulted in AttributeError.

EVA.set_extremes` hotfix for calculating default block size for block maxima and default threshold for peaks-over-threshold.

Added an option to create `EVA` model from extremes directly via `from_extremes` method. This method can be used in situations when source data is not available and only extremes (e.g. annual maxima) are known.

Another new method `set_extremes` allows to set extreme values directly without using the `get_extremes` method. This method can be used to analyze extreme values extracted using a custom method or with/without specific values (e.g. outliers).

Changes in this release:

- Performance improvements due to delayed evaluation of log messages via string interpolation.
- Changed the default block size - changed from "1Y" to "365.2425D". This was due to `pandas` deprecating the "Y" time unit and because using days removes ambiguity regarding the definition of a year.
- Added [extras](https://packaging.python.org/tutorials/installing-packages/#installing-setuptools-extras) to pip-based installation.
- Added an option to the Block Maxima extremes extraction to discard the last block if its duration is smaller than `min_last_block`, which is a value in the range [0;1)