Update gradient logic for Qiskit Rust circuit data implementation

[woodsp-ibm]

Summary

Fixes #181

This updates internals of the gradients which more directly manipulated circuit data the way it was when it was managed in Python. The Rust implementation is changing things to be more performant but things are also different so this changes the logic to try accommodate as things are currently done, which is what is in Qiskit stable releases at present, and the new Rust implementation which is presently in main branch of Qsikit repo. #181 notes the aspect of CI which is failing is when we test against main which was done to monitor for any problems here due to changes.

This has partially fixed things but there is another aspect, that I failed to notice/note in #181 that also occurs

      File "/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/qiskit/circuit/parameterexpression.py", line 418, in __float__
    raise TypeError(

    TypeError: ParameterExpression with unbound parameters (dict_keys([Parameter(__gθ0)])) cannot be cast to a float.

which has not been addressed at present.

Details and comments

[woodsp-ibm]

While this would be mergeable, as the tests against main, though they fail do not prevent things from being merged. I labelled this PR onHold as it would be better to have the remaining problem, as noted above, sorted so things pass completely against main.

The remaining problem, i.e.this failure

TypeError: ParameterExpression with unbound parameters (dict_keys([Parameter(__gθ0)])) cannot be cast to a float.

ends up being around the Rust implementation and from these notes, that talk about potential issues, https://github.com/Qiskit/qiskit/blob/main/releasenotes/notes/circuit-gates-rust-5c6ab6c58f7fd2c9.yaml#L47-L79 I added a workaround that simply assigns the current circuit .data back to itself and this "fixes" the issue. So I will commit that momentarily.

[coveralls]

Pull Request Test Coverage Report for Build 10177248383

Details

• 20 of 23 (86.96%) changed or added relevant lines in 3 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.02%) to 90.444%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit_algorithms/gradients/reverse/derive_circuit.py	16	19	84.21%

Totals
Change from base Build 10020859087:	0.02%
Covered Lines:	6370
Relevant Lines:	7043

---

💛 - Coveralls

[ElePT]

Thanks a lot @woodsp-ibm for keeping the code updated. I was planning to eventually look into it but you got there faster :)

[woodsp-ibm]

The last change, to file gradients/utils.py, is not needed with this PR Qiskit/qiskit/pull/12794, so once this merges to Qiskst main I will remove the workaround from that (i,e the data copy)

[Cryoris]

Qiskit/qiskit#12794 (Jake's PR) fixes parts of this PR, as it removes the need for copying circuit.data. However, it creates new issues due to the fact that derive_circuit uses the private _parameter_table attribute. To avoid such breakage in future, woodsp-ibm#2 rewrites derive_circuit to only use public API. With that PR and Qiskit/qiskit#12794 all tests pass as expected.

So we can either

1. Merge this as is and merge the public API PR as follow-up fix once Jake's PR is released
2. Wait until Jake's PR is released in 1.2 and merge this PR with the PR that changes to using only public API

With Qiskit 1.2 imminent maybe option 2 is better?

[woodsp-ibm]

@Cryoris Well it seems that things need a little fixup having merged your PR. A couple of copyright dates and mypy typing by the looks.

@ElePT It seems with main that more stuff is broke - maybe with 1.2 as well. This PR Qiskit/qiskit#12824 changed base types so computeuncompute now fails e.g.

test.time_evolvers.test_pvqd.TestPVQD.test_zero_parameters
----------------------------------------------------------

Captured traceback:
~~~~~~~~~~~~~~~~~~~
    Traceback (most recent call last):

      File "/home/runner/work/qiskit-algorithms/qiskit-algorithms/test/time_evolvers/test_pvqd.py", line 216, in test_zero_parameters
    fidelity_primitive = ComputeUncompute(sampler)

      File "/home/runner/work/qiskit-algorithms/qiskit-algorithms/qiskit_algorithms/state_fidelities/compute_uncompute.py", line 83, in __init__
    raise ValueError(

    ValueError: The sampler should be an instance of BaseSampler, but got <class 'qiskit.primitives.sampler.Sampler'>

[Cryoris]

Copyright & mypy done in 0c99ccb

[woodsp-ibm]

Sampler base class change reverted in Qiskit/qiskit#12871

[woodsp-ibm]

With Qiskit/qiskit#12794 having been merged earlier today I re-ran the CI and it all passes now. So it looks like this is good to go.

[Cryoris]

LGTM but having touched the code maybe @ElePT should approve too 😄