If you pass an argument to train that is not one of its arguments, it gets passed to the underlying model function (ranger, in this case). You could use something like
train(Species ~ ., data = iris, method = "ranger", always.split.variables = "Sepal.Width")