Leverage Dict-based aux_operators

[mrossinek]

What is the expected enhancement?

Terra has just added support for Dict-based aux_operators in its Eigensolver interfaces (Qiskit/qiskit#6870). This can greatly improve the handling within the property framework.

[mrossinek]

I believe the aforementioned PR also slightly changed the list-based interface by removing one level of list-nesting. This went unnoticed until the CI on the main branch failed last night, presumably because the terra unittests do not cover the aux_operators interface? 🤔

[mrossinek]

Actually, there now appears to be a mismatch between the aux_op_eigenvalues as returned by the NumPyMinimumEigensolver compared to the VQE...

[mrossinek]

Okay some more observations:

• the NumPyEigensolver has (since its first commit) returned the auxiliary operator eigenvalues as a tuple: (value, 0). current code. It is not 100% clear to me, why it is done this way. I also did not find documentation on this matter. Neither in the algorithm itself nor in the Result object. The only suggestion I have would be that the 0 is a place-holder for when we will have standard-deviation information on this eigenvalue (which in the numpy case is always 0).
• in order to match this interface (that each aux_op_eigenvalue is actually a tuple) the VQE used to wrap each evaluated aux_op into a list here which is no longer the case in the new code here

There are two solutions:

1. Remove the tuple style from NumPyMinimumEigensolver (strongly disfavored)
2. Change the VQE to also return tuples (not simply wrap the eigenvalue into a list, imo) and document the return type properly for the future.

[mrossinek]

I removed the critical label again, because this is an issue which needs to be fixed on the Terra side. See my PR: Qiskit/qiskit#7144