Should we implement our transpiler as a Qiskit transpiler plugin?

[jordandsullivan]

At UnitaryCon, I learned about this feature that Qiskit has for implementing custom transpiler passes: https://docs.quantum.ibm.com/guides/transpiler-plugins.
For MVP, this seems like it could be a cleaner way to use a subset of Qiskit functions and objects rather than copy-pasting the entirety of the Qiskit files themselves as we have currently done:

An existing example of this is a TKET-Qiskit compiler plugin (i.e. you can then combine TKET and Qiskit compilation passes).

On the other hand, one reason we wanted to copy/paste the transpiler pass files in their entirety was to avoid depending on a relatively unstable module of Qiskit.

[jordandsullivan]

@sonikaj interested to hear your thoughts on this.

[jordandsullivan]

I would argue we should likely be moving away from tighter integration with Qiskit, as we have already run into multiple breaking changes since beginning development of UCC.