Merge geo-cross-validation to geone

[pjuda]

Key features:

• implements DeesseEstimator class (wrapping DeesseInput) which is compatible with scikit-learn estimators. Therefore it can be used with scikit-learn's cross_validate, GridSearchCV, etc.
• implements cross-validation metrics with a flexible SkillScore class compatible with scikit-learn
• adds gslib module, reading classical "pointset" gslib file to a dictionary of arrays which can be converted to pandas DataFrame. This module is necessary to provide a suitable input for DeesseEstimator,
• gslib module and metrics are tested using python.unittest.

No external dependencies were added.

The next step would be:

• [x] add an executable scipt geone-cv performing cross-validation,
• [x] add DeesseEstimator tests (using scikit-learn's template),
• [x] add jupyter example for cross-validaiton,
• [x] document cv_metrics,
• [x] add requirements.txt reproducing environment for testing and jupyter examples.

EDIT: the tasks above are finished

[pjuda]

The tasks above are finished. I think that geone-cv is ready to be merged.

Further steps, before submitting the cross-validation paper:

• make the project open-source, when Julien is ready,
• label the latest version,
• get a DOI on zenodo.

[pjuda]

I'm closing the pull request for the moment as I'm not sure if I could improve further the example use case. Now the user needs to build the DeesseEstimator from scratch (by copying their DeeSseInput...). Maybe it's better to reuse the DeesseInput to construct the estimator. In this way, the cross-validation can be more easily applied to existing models without the need to copy existing code.