Add integer support to nearest neighbor interpolation

```
RuntimeError: Expected tensor for argument #1 'weight' to have one of the following scalar types: Half, BFloat16, Float, Double; but got torch.IntTensor instead (while checking arguments for embedding_bag)
```