Make use of AerPauliExpectation for gradients, quantum Fisher information

[adekusar-drl]

What is the expected enhancement?

Make sure AerPauliExpectation can be used for gradients, QFI, etc

[Cryoris]

While the code allows to use AerPauliExpectation, it currently gives wrong results due to a bug in the expectation values if coefficients in the state function are involved -- as is the case for gradients.

[Cryoris]

With Qiskit/qiskit#6497 merged, this issue is resolved.

The AerPauliExpectation now yields the correct values, see this plot which shows the absolute difference between gradients calculated exactly (using the matrix expectation) again Pauli and AerPauli:

For the QFI, the errors look very similar.

Here's a small runtime comparison for gradients (not including matrix expectation because that scales way too bad and reaches the 40s mark already for 12 qubits):