Summary

When converting from a DAGCircuit to a DAGDependency, in some cases it would be useful to compute and return additional information related to this conversion, such as the mappings between the input DAGOpNodes and the output DAGDepNodes. In particular, this would allow analysis passes that translate a DAGCircuit to a DAGDependency, collect blocks of nodes from the DAGDependency (exploiting the commutativity information offered by the latter structure), find the lists of corresponding nodes in the origin DAGCircuit and (for example) resynthesize these nodes. A direct application would be to extend the new StarPreRouting pass to collect and resynthesize "star-shaped" pieces of the circuit while also exploiting the commutativity between the nodes.

This PR adds a new class CircuitConversionData that stores the mappings and extends the conversion functions to dag_to_dagdependency_with_data and _dag_to_dagdependency_v2_with_data (all of the code pertaining to DAGDependency_V2 is private for now).

The reason that CircuitConversionData is a class rather than a mapping is that we might want to store more information in the future, and we would not need to keep extending the API. And it would also probably be easier to convert this to Rust in the future.

I am wondering whether I should make the class CircuitConversionData also private or also extend other circuit converters such as circuit_to_dagcircuit, etc. What do you think?