Pyextremes

This update fixes a bug in the `plot_return_value_stability` function and adds support for multiple distributions in the `plot_return_value_stability` function.

This is a big update involving major refactoring of the code base. Nevertheless, API mostly hasn't changed since the last version - only small things like support attribute and property names were changed to more properly reflect their purpose. However, because some changes were made to the attribute names people may be using a decision was made to release this as version 2.0.0. Overall, the following notable changes were made to the code:

- `emcee` version 2 is no longer supported - there are major differences with the way MCMC chains are stored between versions 2 and 3; because of this only version 3+ will be supported moving forward
- when using the `EVA` class and when no distribution is specified by the user the model now automatically selects the best distribution using the Akaike information criterion (AIC)
- MLE model is now significantly faster when calculating confidence intervals for large number of samples - this was mostly achieved by using the built-in `multiprocessing` module
- results for all models are now cached as a simple dictionary in the `.return_value_cache` attribute; no more unnecessary complexity with self-made caching
- confidence intervals are no longer plotted by default - this change was made in order to increase performance for the default case

This is the first, fully functional release of the `pyextremes` library. All features are working and have been tested. The two main changes relative to the previous release **v0.2.0** have been made:

- package has been extensively tested in production with real-life projects and data, multiple bugs and inconveniences have been fixed, some plots and outputs were updated to match industry standards
- added plotting functions assisting during threshold selection for the POT-GPD family of models

This is a fully working, nearly finished release of `pyextremes`. This release includes all major tools envisioned for the package and has been tested against most scenarios.

These are the changes relative to the `v0.1.0` release:

- all models are now based on the same universal `Distribution` class. This class is a wrapper around `scipy.stats.rv_continuous` and allows to use any `scipy.stats` distribution in the models and to create custom user distributions subclassed from `scipy.stats.rv_continuous`
- fully reworked how models work - all models now have most of their functionality defined in the base class, with only the `fit`, `encode_kwargs`, and `_get_extreme_value` methods defined for each model

This is an alpha release of the `pyextremes` library. It is possible to do the following operations:

- extract extreme values and plot them
- fit a model
- get return values for given return periods

There is still a lot to be done, but for basic purposes of this project - it's ready to be used.