Canonicalization of inverse extract/insert pairs on the same index is obfuscated if there is a chain of inserts

Consider the circuit CNOT(0, 1) - X(0)

  %0 = quantum.alloc( 2) : !quantum.reg

  %1 = quantum.extract %0[ 0] : !quantum.reg -> !quantum.bit
  %2 = quantum.extract %0[ 1] : !quantum.reg -> !quantum.bit
  %out_qubits:2 = quantum.custom "CNOT"() %1, %2 : !quantum.bit, !quantum.bit
  %3 = quantum.insert %0[ 0], %out_qubits#0 : !quantum.reg, !quantum.bit
  %4 = quantum.insert %3[ 1], %out_qubits#1 : !quantum.reg, !quantum.bit

  %5 = quantum.extract %4[ 0] : !quantum.reg -> !quantum.bit
  %out_qubits_0 = quantum.custom "PauliX"() %5 : !quantum.bit
  %6 = quantum.insert %4[ 0], %out_qubits_0 : !quantum.reg, !quantum.bit

  quantum.dealloc %6 : !quantum.reg

After --canonicalize, the IR stays unchanged.

Proposed improvement:
By interchanging the order of the two inserts after the CNOT:

  %0 = quantum.alloc( 2) : !quantum.reg

  %1 = quantum.extract %0[ 0] : !quantum.reg -> !quantum.bit
  %2 = quantum.extract %0[ 1] : !quantum.reg -> !quantum.bit
  %out_qubits:2 = quantum.custom "CNOT"() %1, %2 : !quantum.bit, !quantum.bit
  %4 = quantum.insert %0[ 1], %out_qubits#1 : !quantum.reg, !quantum.bit
  %3 = quantum.insert %4[ 0], %out_qubits#0 : !quantum.reg, !quantum.bit

  %5 = quantum.extract %3[ 0] : !quantum.reg -> !quantum.bit
  %out_qubits_0 = quantum.custom "PauliX"() %5 : !quantum.bit
  %6 = quantum.insert %4[ 0], %out_qubits_0 : !quantum.reg, !quantum.bit

  quantum.dealloc %6 : !quantum.reg

we should be able to detect the inverse pair on index 0:

  %0 = quantum.alloc( 2) : !quantum.reg
  %1 = quantum.extract %0[ 0] : !quantum.reg -> !quantum.bit
  %2 = quantum.extract %0[ 1] : !quantum.reg -> !quantum.bit
  %out_qubits:2 = quantum.custom "CNOT"() %1, %2 : !quantum.bit, !quantum.bit
  %3 = quantum.insert %0[ 1], %out_qubits#1 : !quantum.reg, !quantum.bit

  %out_qubits_0 = quantum.custom "PauliX"() %out_qubits#0 : !quantum.bit
  %4 = quantum.insert %3[ 0], %out_qubits_0 : !quantum.reg, !quantum.bit
  quantum.dealloc %4 : !quantum.reg

Start by reproducing the supplied MLIR circuit and running --canonicalize, then trace the canonicalization handling for quantum.extract and quantum.insert through the chained inserts. Done means the inverse pair at index 0 is detected and the IR is reduced to the form shown in the proposed improvement.