Update our use of `torch.meshgrid` due to pytorch changes

Hi @ppwwyyxx ,

## Instructions To Reproduce the Issue:

I was able to install detectron2 in windows 10 and was also able to build and create exe (torchscript_mask_rcnn.exe) file.

The model.ts file was downloaded using the commands provided in 'https://github.com/facebookresearch/detectron2/tree/main/tools/deploy' . But when executing the exe file with downloaded models, it fails at step
```
// Run the network
auto output = module.forward({inputs});
```
without any error message. It gave below warning message.
```
[W ..\..\aten\src\ATen\native\TensorShape.cpp:2157] Warning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (function operator ())'
```
I tried for both 'scripting' and 'tracing'. Can you please advise. Thanks.

## Environment:

CUDA : 10.2
torch==1.10.0
torchvision==0.11.1