Enable passing a numpy.random.RandomState object as random_state_model in the train functions.

[mcw92]

Currently, the train functions only accept integers as the model random state in random_state_model as it makes the internal handling of the different RandomState objects eventually required for different ranks in the DistributedRandomForest class easier.
For the data-related random state, each rank needs the same seed / RandomState object which makes it easy to further process a RandomState object passed here. For the model, the original workflow took the model base seed, added the local rank, and turned it into a rank-specific RandomState object which would not work the same if a RandomState object would be passed here. If we would allow passing a RandomState object for the model as well, it could be used to generate either the model base seed (in the case of local model seed = model base seed + rank) or the sequence of comm.size integers used as local seeds.