Implement a QNN based on the Sampler primitive #400
Description
What should we add?
Terra has implemented new ways to run circuits that may potentially speed up QNN training in the cloud. In this issue we implement a new type of QNN, a SamplerQNN that will be based on the Sampler primitive.
The implementation should be given an instance of the Sampler instead of a quantum circuit and should re-use the sampler for every evaluation of the circuit.
The implementation should be tested again the reference implementation in Terra and cloud versions of the primitives. The latter to be decided how to implement.