Add external dependencies for training strategies

[MartinuzziFrancesco]

Packages like LIBSVM or MLJLinearModels can be external (weak) dependencies.

Add a lower level function that takes the place of _train structured like train(algo::TrainingAlgo, states, target; kwargs...). The upper level function still maintains train(esn::AbstractReservoirComputer, target_data, training_algo; kwargs...). Additionally include a version that has train(esn::AbstractReservoirComputer, training_data, target_data, training_algo; kwargs...) to prepare for #202