Optimize the evaluation of "deep" categorical fields

[vruusmann]

The "depth" of a categorical field can be estimated as the number of Value child elements inside the parent DataField or DerivedField element. For example, a DataField element with ten Value child elements might be considered "shallow", and a DataField element with more than one thousand Value child elements might be considered "deep".

The "depth" information can be used to optimize the execution plan for several model types. For example, the evaluation of the RegressionModel element currently involves iteration over all the entries of the RegressionTable element. Such "linear scan" approach is highly inefficient in case of "deep" categorical fields, and should be replaced with "dictionary lookup" approach. By definition, this will result in performance increase, which is proportional to the "depth" of the categorical features at hand.