Smote-variants

* vectorized implementations for most of the techniques to boost performance
* a refactored and improved evaluation and model selection toolkit
* 100% test coverage
* 10.0 PEP8 conformancy (by pylint)
* polynom_fit_SMOTE split into 4 different techniques

In the 0.5.1 release the package description was corrected in addition to the changes in the 0.5.0 release:

A decent refactoring was carried out by splitting the individual oversamplers to separate files.
SYMPROD added as the 86th oversampler implemented, thanks to intouchkun
An experimental feature: metric learning to determine the metric to be used for neighborhood estimation added to 69 oversamplers.
A sample notebook to illustrate metric learning backed oversampling was added.
Support and CI pipelines for Python 3.9 and 3.10 added.

In the 0.5.0 release:

1) A decent refactoring was carried out by splitting the individual oversamplers to separate files.
2) SYMPROD added as the 86th oversampler implemented, thanks to intouchkun
3) An experimental feature: metric learning to determine the metric to be used for neighborhood estimation added to 69 oversamplers.
4) A sample notebook to illustrate metric learning backed oversampling was added.
5) Support and CI pipelines for Python 3.9 and 3.10 added.

- code refactored to fulfill PEP8 standards
- minor bug fixes

This patch fixes some pickle dumping performance issues.