Whole-history-rating

Everything in config is saved unless something is unpickable (in that case only w2 debug and uncased are saved)

Enhanced docstring readability for improved code documentation.
Update readme.
Refactored `load_games` function.
Removed the multiplication by 100 of the uncertainty coefficient in the player ratings retrieval process.
Replaced the `inspect` method in the `Game` class with the `__str__` method for more intuitive object representation.

**new**:
- add proper tests
- use black formatting
- add utils script to contain error classes and utility functions

**updated**:
- modify inspect so that it’s coherent across save and load
- modify playerday elo to a property attribute
- modify player log_likelihood computation to comply with tokoharu pull request
- modify player hessian, covariance and update_by_ndim_newton computation to comply with tokoharu pull request

+add possibility to set the base to uncased (shusaku and Shusaku will be the same player)
+you can get ordered ratings both in compact form and only current elo
~you can put 'b' or 'w' in lower case for the winner

delete two prints in _setup_game and create_game