@Patschkowski pointed out in #3909 that using KFoldCV with RandomForest<> did not seem to work. I thought it would be simple to figure out why but I actually had to dive in fairly deep. As it turned out the fix was simple:

• KFoldCV uses the class MetaInfoExtractor, which determines various information about a given machine learning algorithm, like what Train() variants it supports and so forth.

• MetaInfoExtractor, in order to determine what variants are available, uses the HAS_METHOD_FORM() macro in sfinae_utility.hpp, which allows a method to either meet a fixed given form, or have additional extra arguments.

• The number of extra arguments is limited to 7... but RandomForest::Train() has 8 extra hyperparameters after the training data and labels! Therefore MetaInfoExtractor does not work.

• The solution is to increase the maximum number of extra allowed arguments to 10.

• I then added tests to cv_test.cpp to ensure that RandomForest<> works correctly both with MetaInfoExtractor and with KFoldCV.

• I also noticed that the labels and weights types were not templatized for RandomForest<>, so I also generalized those.