Gradient dialect removes `qnode` attribute from quantum functions when creating helper functions.

• Expected behavior: The attribute remains through the lowering of gradient and quantum dialect up until LLVM-IR (in which case we might wish to annotate it with some metadata).

• Actual behavior: The gradient transforms generates helper functions which will strip the qnode attribute from qnodes. E.g.,

  func.func private @f(%arg0: tensor<f64>) -> tensor<f64> attributes {diff_method = "parameter-shift", llvm.linkage = #llvm.linkage<internal>, qnode} {
  }

Changes to:

  func.func private @f.quantum(%arg0: memref<f64>, %arg1: memref<?xf64>) -> memref<f64> attributes {gradient.qgrad = @f.qgrad, passthrough = ["noinline"]} {
}

This causes async to not be triggered since async looks for the qnode attribute.

• Reproduces how often: Always

No file or test is named. Trace the gradient and quantum dialect lowering that creates the helper functions shown in the LLVM-IR example, then follow how function attributes are copied or dropped. Confirm that qnode remains available through lowering so async can still detect it, and add coverage for the reproduced case if the existing tests provide a suitable location.