`decompose-lowering`'s `prepareCallOperands` doesn't support qreg + multiple individual qubit indices

Signatures of the form (qreg, *params, first_qubit_index, second_qubit_index, ...) are not supported by decompose lowering.

MWE:

@qjit(capture=True)
@qnode(device("lightning.qubit", wires=2))
def circuit(wires):
    qp.PauliRot(0.2, "XZ", wires)
    return qp.probs()


circuit([0, 1])
print(circuit.mlir)

generates the IR

    func.func public @circuit(%arg0: tensor<i64>, %arg1: tensor<i64>) -> tensor<4xf64> attributes {diff_method = "parameter-shift", llvm.linkage = #llvm.linkage<internal>, quantum.node} {
      %cst = arith.constant 2.000000e-01 : f64
      %c0_i64 = arith.constant 0 : i64
      quantum.device shots(%c0_i64) ["/Users/river.mccubbin/Github/catalyst/.venv/lib/python3.14/site-packages/pennylane_lightning/liblightning_qubit_catalyst.dylib", "LightningSimulator", "{'mcmc': False, 'num_burnin': 0, 'kernel_name': None}"]
      %0 = quantum.alloc( 2) : !quantum.reg
      %extracted = tensor.extract %arg0[] : tensor<i64>
      %1 = quantum.extract %0[%extracted] : !quantum.reg -> !quantum.bit
      %extracted_0 = tensor.extract %arg1[] : tensor<i64>
      %2 = quantum.extract %0[%extracted_0] : !quantum.reg -> !quantum.bit
      %out_qubits:2 = quantum.paulirot ["X", "Z"](%cst) %1, %2 : !quantum.bit, !quantum.bit
      %extracted_1 = tensor.extract %arg0[] : tensor<i64>
      %3 = quantum.insert %0[%extracted_1], %out_qubits#0 : !quantum.reg, !quantum.bit
      %extracted_2 = tensor.extract %arg1[] : tensor<i64>
      %4 = quantum.insert %3[%extracted_2], %out_qubits#1 : !quantum.reg, !quantum.bit
      %5 = quantum.compbasis qreg %4 : !quantum.obs
      %6 = quantum.probs %5 : tensor<4xf64>
      quantum.dealloc %4 : !quantum.reg
      quantum.device_release
      return %6 : tensor<4xf64>
    }

which fails with decompose-lowering:
Assertion failed: (idx < size()), function operator[], file SmallVector.h, line 295.
because of the assumption that qubit indices with qregs will be represented as a tensor (eg. tensor<2xi64>).

Start in decompose-lowering at prepareCallOperands and reproduce the supplied qjit capture example with a qreg plus individual qubit indices. Trace the handling of qreg index operands and verify that the example lowers without the SmallVector assertion and produces valid MLIR.