Validate input when setting the QuantumCircuit data attribute directly

Similar to #2170 , but for QuantumCircuit. The possibility of arbitrary data in circuit.data makes modifying or replacing the internals of QuantumCircuit difficult and error prone (see discussion around circuit._append in #2824 ). We should:

• Move circuit.data to circuit._data
• Add a getter for circuit.data and document it as a ordered list of (Instruction, qargs, cargs). It’s an open question whether or not we should commit to preserving user-insertion order.
• Add a setter for circuit.data with a DeprecationWarning, directing users to e.g. QuantumCircuit.append and the rest of the circuit API.

[quantumjim]

Thanks!

The relevant part of Ignis is in here: the _make_syndrome_graph() method of GraphDecoder. Any incompatible changes in .data() can be expected to cause this test to fail.

[kdk]

Hi @quantumjim , the title of this issue is somewhat outdated (see the two comments preceding yours, and the PR #2838 ). The current plan is for circuit.data() to remain in place with the same list-like interface, but with changes to circuit.data() being validated to make sure they are of the right Instruction, qargs, cargs format. Existing code should continue to work.

Can you link to the example in ignis that works with .data()? It would be handy to have as a test case.