✨ Pull `GPhaseOp`s out of multi-operation `ctrl` modifiers

[denialhaag]

Description

#1751 has dropped the condition that modifier bodies can only contain a single unitary operation followed by a yield operation.

We should add a canonicalization that pulls out GPhaseOps from multi-operation ctrl modifiers. For example,

%const = arith.constant 0.123 : f64
qc.ctrl(%c0, %c1) targets(%t0, %t1) {
  qc.gphase(%const)
  qc.rxx(%const) %t0, %t1 : !qc.qubit, !qc.qubit
}

is unrolled to

%const = arith.constant 0.123 : f64
qc.ctrl(%c0) targets(%c1) {
  qc.p(%const) %c1 : !qc.qubit
}
qc.ctrl(%c0, %c1) targets(%t0, %t1) {
  qc.rxx(%const) %t0, %t1 : !qc.qubit, !qc.qubit
}

Notice that we already have such a canonicalization for one-operation ctrl modifiers. It should be possible to adapt some of the existing code.

Acceptance criteria

• The described canonicalization is implemented in QC and QCO
• The two comments made in #1603 are taken into account

[burgholzer]

Note that the example is not quite correct (at the time of posting this), as the target for the controlled phase gate should be the second control and not the first target.

[denialhaag]

Note that the example is not quite correct (at the time of posting this), as the target for the controlled phase gate should be the second control and not the first target.

Ohh, my bad! Fixed!