Improve backend circuit execution

[woodsp-ibm]

Goal is to "modernise" QuantumInstance and run_circuits logic for execution of circuits on the backend given advances in capability of Terra and IBM Q Provider since some this aforementioned logic was written. This includes areas such as

1. Moving over to use of the ibm q job manager when the remote provider is IBM Q.
2. Ensuring we work with other remote providers correctly general question related to AQT simulator and Qiskit. "Attribute error" #5706
3. More reliable operation with backend Qiskit/qiskit-aqua#910
4. And in conjunction with 3 above perhaps looking to allow a job to be cancelled and resubmitted on behalf of an algorithm when the job seems to be stuck. Not sure how much this is catered to in 1 above. But since a variational algo can send in many jobs having it get stuck as it progress and having to restart the entire algo is less than desirable. Auto-recovery needs to be looked at but maybe it (also) can be done in conjunction with some user supplied code in a the job monitor callback which would signal to cancel/retry.
5. Algorithms can require multiple circuits; since there is a backend limit on experiments per job currently they are split into multiple jobs if the limit is exceeded. Some like QuantumKernel do a circuit for each data pair and group by a batch size. If the grouping is more for execution efficiency rather than algorithmic reasons it might be better to inform such grouping via the QI rather than expect an end user to know things like max experiments and have to configure things manually (in QK case via some batch size param)

These are related to improving execution etc too

• Ability to stop job if it fails with 9999 error code #5720
• Algorithm may hit job limit - tone down emitted message #5765

[woodsp-ibm]

Algorithms have been refactored over to use primitives and circuit execution etc is now handled by these. As such I am closing this issue.