Allow to pass callables as outer CV (#90)

Passing an unsized iterable (such as a generator) as an outer CV does not work if the experiment has several datasets/estimators. as in that case the iterable is exhausted in the first iteration, raising
`StopIteration` for the following datasets/estimators.

We can instead allow to pass a callable that returns a fresh instance of the iterable for each combination.

Note: the type hints and documentation have not been updated for now.

Author: vnmabus

skdatasets/utils/experiment.py

@@ -100,6 +100,7 @@ def get_n_splits(
 CVLike = Union[
     CVSplitter,
     Iterable[IndicesType],
+    Callable[[], Iterable[IndicesType]],
     int,
     None,
 ]
@@ -176,6 +177,9 @@ def _iterate_outer_cv(
     y: TargetType,
 ) -> Iterable[Tuple[DataType, TargetType, DataType, TargetType]]:
     """Iterate over multiple partitions."""
+    if callable(outer_cv):
+        outer_cv = outer_cv()
+    
     if isinstance(outer_cv, Iterable):
         outer_cv, cv_copy = itertools.tee(outer_cv)
         if len(next(cv_copy)) == 4: