Wildcard: allow running our notebooks with simulators?

[Eric-Arellano]

We are using more hardware rather than simulators in our notebooks. This is meant to help users by showing real-world workflows, where the emphasis is running on actual hardware rather than simulators.

We don't want to complicate the notebooks by showing both simulators and hardware.

However, that poses a challenge for our docs writing pipeline:

1. We cannot easily test the jobs in every CI run (Test the safe parts of notebooks that run on hardware in every CI run #1173)
2. It is harder for us to get sample output for the notebooks to show to the user

This issue is to explore the idea of allowing us to dynamically rewrite our notebooks to use simulators rather than hardware when using our tool nb-tester. It would, for example, rewrite the backend to use FakeManilaV2. We'd need to figure out the rewriting of the code, along with reckoning with simulators not working when the qubit count is too high.

[frankharkins]

@beckykd had the idea of overwriting the least_busy function to always return FakeManilaV2. I'm trying it out now with a patched fork and it seems to work pretty well bar a few quirks we can probably iron out.

1. docs/start/hello-world.ipynb fails in the final cell with

   AttributeError: 'DataBin' object has no attribute 'ensemble_standard_error'

   Because the fake backend doesn't set that attribute. We could just avoid using this attribute as they're only used for error bars in the plot.

2. tutorials/submitting-transpiled-circuits/notebook.ipynb fails when using Batch. It's possible this notebook is just broken anyway as there were some changes to batch mode recently.

   AttributeError: 'QiskitRuntimeLocalService' object has no attribute 'backend'

3. This doesn't work for tutorials/build-repitition-codes/notebook.ipynb and tutorials/build-repitition-codes/notebook.ipynb as they use qiskit-ibm-provider. We probably want to switch them to runtime anyway.

Here's the patched function: https://github.com/frankharkins/qiskit-ibm-runtime/blob/fake-runtime/qiskit_ibm_runtime/qiskit_runtime_service.py#L1108-L1132

---

UPDATE: I tried this out on #1157 and it gets a bit hairier.

TranspilerError: "The control-flow construct 'if_else' is not supported by the backend."

Maybe we still run some notebooks in the fortnightly cron job, but the well-behaved notebooks can use the patched package?

[Eric-Arellano]

@frankharkins it may be desirable for us to always try to have the notebooks work without issue with local simulators, and even communicate to users that they can swap out .least_busy() with FakeManilaV2. That allows us to have hardware-first docs, but that are still accessible. So, when we have issues like the above, that means we should update the docs to work with simulators. The one fundamental barrier I see is if we need to use utility-scale # of qubits?

Re: implementation, we could maybe have nb-tester add a code block to "monkeypatch" Runtime .least_busy() for us, but hide it from users. Python is quite flexible like that:

>>> type(str("123"))
<class 'str'>

>>> str = int
>>> type(str("123"))
<class 'int'>