Add RandomTreesEmbedding feature

[kulbachcedric]

Hi @MaxHalford,
if I understood the RamdomTreeEmbeddings correctly, this PR should implement them.
The crucial lines are the following, where the transforms_one methods returns a dictionary showing which leaf x ends up in for each tree.

features = {}
for tree_id, tree in enumerate(self.trees):
    leaf = tree if isinstance(tree, HSTLeaf) else tree.traverse(x)
    features[(tree_id, self._leaf_indices[tree_id][id(leaf)])] = 1.0
return features 

Could you please briefly check it ?

[MaxHalford]

Nice! The one thing that's missing is that the implementation assumes features are scaled between [0, 1]. We need to document in every place where this assumption is made that the user should use a MinMaxScaler prior. See what I mean?