✨ Add rotation-through-CX optimization pass

Problem Statement

Following up on #1107 and inspired by the plenty of optimization passes Qiskit provides (many of which will soon be covered in the Predictor), I came across Optimize1qGatesSimpleCommutation, which (among other things) does:

RZ(a)──CX(control)──RZ(b)  →  CX(control)──RZ(a+b)
RX(a)──CX(target)──RX(b)   →  CX(target)──RX(a+b)

Current adjacent-rotation canonicalization and merge-single-qubit-rotation-gates cannot cross a CX. Target-level two-qubit fusion also does not cover dynamic angles or a lone CX.

Proposed Solution

I propose implementing this as a lightweight pass that runs before merge-single-qubit-rotation-gates.

A later iteration could move toward parity with Qiskit’s Optimize1qGatesSimpleCommutation by supporting P on the CX control, X/SX on the target, RZX blockers, broader one-qubit-run resynthesis, target-aware profitability, and fixed-point execution.

Wrong-wire rotations, RY, multi-controlled X, and movement across region or control-flow boundaries could be separate generalizations.

Locate the existing adjacent-rotation canonicalization, merge-single-qubit-rotation-gates, and target-level two-qubit fusion passes, then inspect how passes are registered and ordered. Implement the lightweight pass before merge-single-qubit-rotation-gates and verify the stated RZ/control and RX/target transformations while preserving the listed blockers and region or control-flow boundaries.