Cast error with PyTorch 1.12 in label

Hello,

Thanks for making your code available.

Since the update to PyTorch 1.12, I have an error due to a float label value not castable to a long one:

```
File "[...]/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 49, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "[...]/insightface/recognition/arcface_torch/dataset.py", line 166, in __getitem__
label = torch.tensor(label, dtype=torch.long)
TypeError: 'float' object cannot be interpreted as an integer
```

https://github.com/deepinsight/insightface/blob/f488f4de5622c4001b52f98845f6ba422c34ec77/recognition/arcface_torch/dataset.py#L165

I tried to regenerate the `.rec` file with labels of type long (by removing `.000000` in the `train.lst`) but that does not work.

However, updating the line with `label = torch.tensor(int(label), dtype=torch.long)` solves the issue.

Does anyone get the same problem?

Thanks.