[EXPERIMENTAL] Part 1: The new Pass infrastructure

[hamidelmaazouz]

The original PR was deemed too complex and we decided to split it in three main parts. The first part is the pass infrastructure, the second part is control-flow infrastructure, and the third part leverages previous work for a faster Qblox integration. This PR covers only the first part.

• For lack of a proper IR, the new pass infrastructure treats the InstructionBuilder as a (linear) "IR". This is the reason why IR in the pass_base module is intentionally left opaque because there will be more work either improving the existing instruction builder or coming up with a new IR.
• Although the new pass infrastructure is in large inspired from LLVM's new PassManager, most complexities (especially around pass result caching and invalidation) have been stripped out. Great effort has been invested to keep the implementation details simple and in tune with today's rather trivial needs.
• We hope that technical notes in the doc strings help contributors understand decision rationales as well as improve on the design where features can be easily added as requirements arise.
• A particular highlight is the "concept-based polymorphism" as it's featured for passes and analysis results. Although it is largely associated with C++ applications, we saw great benefit in trying to leverage its essence for this design given Python's dynamic typing.
• The proposed "concept-based polymorphism" would therefore not fully honour the original idea but it helps avoid unnecessary inheritance-based polymorphism and fosters a flat and plugin-like paradigm where passes can be easily registered and can individually and thoroughly be tested.
• XEngine.optimize() and XEngine.validate() become a sequence of transform and validation passes, some Control Hardware related verifications can be written as a pass: no need to introduce ad-hoc inheritance on the engine (see NewQuantumRuntime to see how dynamic dispatch is avoided).

[hamidelmaazouz]

Hi @alillistone-OQC. Whenever I open a PR from my fork, this failure arises because the action could not find a branch / revision on the OSS repository. One workaround is to create a duplicate branch on the OSS repository. While this works, it's burdensome and I'm wondering if you could help me fix it once and for all ^^.

[willh727-OQC]

🥳 This needs to happen! But also want to see it act on Control flow so we can handle the complex of building QBlocs and ArbSeq instructions so proper analysis can occur.