[TorchOnnxToTorch] Generate unexpected unflatten for ScatterND

[vfdff]

• reproduce command: torch-mlir-opt -convert-torch-onnx-to-torch scatternd.mlir

  func.func @test_scatternd(
      %data: !torch.vtensor<[18,8,64,64],f32>,
      %indices: !torch.vtensor<[18,4,2],si64>,
      %updates: !torch.vtensor<[18,4,64,64],f32>
  ) -> !torch.vtensor<[18,8,64,64],f32>
  attributes {torch.onnx_meta.opset_version = 13 : si64} {
    %0 = torch.operator "onnx.ScatterND"(%data, %indices, %updates)
      : (!torch.vtensor<[18,8,64,64],f32>, !torch.vtensor<[18,4,2],si64>, !torch.vtensor<[18,4,64,64],f32>)
      -> !torch.vtensor<[18,8,64,64],f32>
    return %0 : !torch.vtensor<[18,8,64,64],f32>
  }

• Now it generate following IR for last unflatten, whose indices is [18,8,64,64], so it is wrong （record issue 4686), and we expect the %25 = torch.prim.ListConstruct %0, %1 : (!torch.int, !torch.int) -> !torch.list

    %25 = torch.prim.ListConstruct %0, %1, %2, %3 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
    %26 = torch.aten.unflatten.int %24, %int0_0, %25 : !torch.vtensor<[144,64,64],f32>, !torch.int, !torch.list<int> -> !torch.vtensor<[18,8,64,64],f32>

• the related code is ScatterND, the dataDims are the full shape of the unflatten, but we expect it is the shape related to the dim0 only, maybe something like unflattenIndicesDims

SmallVector<Value> unflattenIndicesDims;
for (int64_t i = 0; i < indicesLastDim; ++i) {
   unflattenIndicesDims.push_back(dataDims[i]);
}

[vfdff]

I find the MR 4586 will fix this issue