Translate parameterized gates only in gradient calculations

[Cryoris]

Summary

Closes #9054.

Details and comments

All gradient now preserve unparameterized operations instead of attempting to unroll them.
This

• allows to evaluate gradients on custom, opaque gates that individual primitives can handle
• keeps a higher level of abstraction for optimized synthesis and compilation after the gradient circuits have been constructed.

[qiskit-bot]

Thank you for opening a new pull request.

Before your PR can be merged it will first need to pass continuous integration tests and be reviewed. Sometimes the review process can be slow, so please be patient.

While you're waiting, please feel free to review other open PRs. While only a subset of people are authorized to approve pull requests for merging, everyone is encouraged to review open pull requests. Doing reviews helps reduce the burden on the core team and helps make the project's code better for everyone.

One or more of the the following people are requested to review this:

• @Qiskit/terra-core
• @ajavadia
• @ikkoham
• @levbishop
• @manoelmarques
• @t-imamichi
• @woodsp-ibm

[Cryoris]

Tagging @a-matsuo for review 🙂

[coveralls]

Pull Request Test Coverage Report for Build 3750990070

• 61 of 63 (96.83%) changed or added relevant lines in 6 files are covered.
• 36 unchanged lines in 4 files lost coverage.
• Overall coverage increased (+0.1%) to 84.64%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/algorithms/gradients/utils.py	4	6	66.67%

Files with Coverage Reduction	New Missed Lines	%
qiskit/extensions/quantum_initializer/squ.py	2	79.78%
src/sabre_swap/layer.rs	2	98.95%
qiskit/opflow/converters/circuit_sampler.py	6	95.57%
qiskit/utils/quantum_instance.py	26	73.11%

Totals
Change from base Build 3747694715:	0.1%
Covered Lines:	64130
Relevant Lines:	75768

---

💛 - Coveralls

[a-matsuo]

It looks good for me except which directory we should put the code as @t-imamichi mentioned.

[Cryoris]

FYI in d567074 I also removed trailing classical registers that weren't used in the LCU gradients (+ QFI) if the add_measurement is False. With the current estimators that's not really a problem but if someone develops their own primitives this could fail if they don't support classical registers.

[a-matsuo]

Thank you @Cryoris LGTM!

[mtreinish]

I only looked at the transpiler pass, but I left a few comments inline. The biggest thing for me is we should add Target support from the start to this pass, especially because the BasisTranslator which is used internally already supports working with the Target.

[mtreinish]

How is this different than the unroll* passes which effectively do this already? Looking at the code it's because this only unrolls parameterized gates, but if we adding another pass to do this we should probably call out why you'd want to use this instead of any of those.

The other related question is whether we want to add this functionality to an existing pass.

[Cryoris]

Good point, it's unrolling plus translating if the unrolling didn't reach the target base. If it were only unrolling then we could add a criterion function or something to the Unroller which we could set to detect parameterized gates. Then we could afterwards do a translation pass on the gates which are not yet in the right basis. So this pass here would be something like

def is_parameterized(gate):
    # returns True if ``gate`` is parameterized, else False

this_pass = PassManager([
   Unroller(basis_gates, criterion=is_parameterized),
   BasisTranslator(...)
])

That would of course mean that we have to construct this pass manually, but since this is used often in gradient calculations I thought it might be generally useful -- but we can also do the above!

[mtreinish]

Thanks for adding the target support, just two small inline comments about it

[mtreinish]

I just left one question inline, but I had a question inline and I'm not sure how you want to proceed with regards to that. I'm fine either keeping the qubit restriction on the target lookup or dropping it depend on the expectations around the pass. So I'm karking this as approved, but if you want to change that line go ahead and feel free to merge this without waiting on a re-approval from me.