torch.nonzero(t, as_tuple=...) does not work with the JIT because the as_tuple signatures are not exposed properly

```
def foo(t):
  return torch.nonzero(t, as_tuple=False)
: RuntimeError: Arguments for call are not valid.
```

This is because of the Python arg parsing done to handle the "as_tuple" kwarg. Note that tracing works fine. 

cc @mruberry @rgommers @heitorschueroff @pmeier @gmagogsfm