"Exact" cost function

[VolodyaCO]

If the backend of the QCBMCostFunction is a simulator, one has access to the exact wavefunction of the quantum circuit. In that case, we shouldn't prepare the circuit and measure in order to estimate the cost function, because we wouldn't use all the available information. We should instead use the exact wavefunction that has the probabilities of each bit string.

Another pitfall of preparing the circuit and measuring is that measurements are not reproducible because most of the times there is not an available random seed. This is very problematic when trying to study circuit training based on this cost function, because it is basically a stochastic value, instead of being deterministic, as it should be for simulators.