[MRG] Implement CircularValidation as a scorer

[YanisLalou]

Fixes #111 Paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4803844&tag=1 Process described in part [5] and [6].

Small thing not in the paper but implemented:

• If the estimator predicts always the same value, then we assume that the backward_estimator will also predicts always the same value.

--> Why? Because for some estimators, if you train them with np.unique(y) = 1 you get ValueErrors (happens with SVC).

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 91.85%. Comparing base (6d59191) to head (6360626).

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #124 +/- ## ========================================== + Coverage 91.72% 91.85% +0.12% ========================================== Files 43 43 Lines 3348 3400 +52 ========================================== + Hits 3071 3123 +52 Misses 277 277 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.