[Feat] Add Gradient Boosted CIF

[Vincent-Maladiere]

• I could reproduce the results of the tutorial part 1 with this updated version of the model
• Need some feedback about the y input of fit. For now, it accepts record arrays, dataframes and dict provided that the dtypes/columns/keys names are ["event", "duration"], and converts everything to a recarray internally.

TODO:

• [ ] add doc
• [ ] add tests
• [ ] add exemples

cc @ogrisel

[ogrisel]

To ease the review and ensure proper testing, let's split into more focussed PRs:

• one for the brier score metrics and its doc + tests
• one for the GB CIF model + its doc + test and examples

The second PR can be a branch based on top of the branch of the first PR.

For the name of the CIF estimator, maybe we could go for GradientBoostingIncidence instead of GradientBoostedCIF to avoid acronyms.

[Vincent-Maladiere]

Closing in favor of #10 and #11