Behaviour of np.generic.real and np.generic.imag is ... really imaginative 

Discovered in https://github.com/numpy/numpy_stubs/pull/14#discussion_r176951593

```python
>>> b = np.bytes_(b'test')
>>> b.real, b.imag
('test', '')

>>> d = np.datetime64(dt.datetime.now())
>>> d.real, d.imag
(numpy.datetime64('2018-03-26T15:22:14.672804'), numpy.datetime64('1970-01-01T00:00:00.000000'))
```

It seems like we should give an error or warning here, in both cases.