indexing with numpy elements

Currently, "the only supported types to index a tensor are integers, slices and torch.ByteTensor". Would it be possible to add numpy arrays as well, especially in the case where they contain only one element? For instance, this code will raise an error:

``` python
a = torch.FloatTensor(3).normal_()
i = np.int32([2, 0, 1])
a[i[0]]
```

And requires to do `a[int(i[0])]` instead.
