Add split statistics option

[mnwright]

So far, save number of observations in each node and the prediction of each node. Example:

rf <- ranger(Species ~ ., iris, num.trees = 5, node.stats = TRUE)
rf$forest$num.samples.nodes
rf$forest$node.predictions

Fixed #541.

[mnwright]

What other statistics do you need? The decrease in impurity, i.e. the split criterion?

[katavuk]

Perfect, thanks for adding the impurity decrease @mnwright!