Scheduling transpiler options in composite experiments

[yaelbh]

T1._set_backend and T2Ramsey._set_backend contain the following lines:

        if not self._backend.configuration().simulator and not isinstance(backend, FakeBackend):
            timing_constraints = getattr(self.transpile_options, "timing_constraints", {})
            if "acquire_alignment" not in timing_constraints:
                timing_constraints["acquire_alignment"] = 16
            scheduling_method = getattr(self.transpile_options, "scheduling_method", "alap")
            self.set_transpile_options(
                timing_constraints=timing_constraints, scheduling_method=scheduling_method
            )

These transpile options are necessary when running delay gates on devices.

When running a parallel experiment that consists of two T1 sub-experiments, these transpile options are missing.

The task:

1. Move the logic that adds these transpile options to the base classes, to a location where the backend is already set and the circuits are known; apply it to any experiment that has circuits with delay gates and runs on a device.
2. Until then, users will have to manually add this line to their code:

exp.set_transpile_options(timing_constraints={"acquire_alignment": 16}, scheduling_method="alap")

This has to be clearly communicated in docstrings and tutorials.

[yaelbh]

@nkanazawa1989 said something about a PR resolving this issue. I'll check with him.

[yaelbh]

@nkanazawa1989 has referred me to #380. To summarize, the issue described here has two different solutions:

1. Reviving parts of Experiment run hook methods #380 - parallel experiments should combine transpiled circuits (each transpiled with its sub-experiment's transpile options), instead of transpiling a combined circuit.
2. The solution written above, of automatically setting transpile options for circuits with delays.

Each of the solutions alone will solve the issue. In my opinion both should be done.

[yaelbh]

@nkanazawa1989 I've presented the idea of sub-experiment transpilations to the Haifa team. They raised a concern that some operations may require a holistic view of the entire circuit. As concrete examples:

1. Consider the case of T1 sub-experiments with different delays. Transpilation of the entire circuit will align them (thanks Haggai for this example).
2. Transpilation introduces ancilla qubits (thanks @ShellyGarion!), even with optimization level 0. A naive composition of two transpiled circuits may result in a circuit that has the union of the ancilla qubits from here and from there.

[yaelbh]

On the other hand, it's not only the transpile options at stake. If a sub-experiment has an implementation for _postprocess_transpiled_circuits then it will not run.

[yaelbh]

Stale, closing