MachineLearning: when using a model, the number of available qubits is determined solely by the number of features

[abacabadabacaba]

In Microsoft.Quantum.MachineLearning, there is a NQubitsRequired function which determines the number of qubits used by a model.

Unfortunately, this function is not used. Instead, only the qubits necessary to encode a sample are made available to the model, and it cannot use any additional qubits.

[guenp]

@cgranade, not sure what the best solution is to this issue aside from a refactor. What are your thoughts?