BUG: skip deprecation for numpy top-level types

[pv]

Deprecation decoration of type objects is more likely to cause problems
than for functions, so don't do it. Deprecation of plain functions is
likely enough to result to deprecation warnings in most cases.

This would likely be important to fix for Scipy 1.4.0, as otherwise we break
the backward compatibility here.

E.g. the array scalar type objects are often used in dtype= arguments
and possibly compared with other scalar objects, and our deprecation
decoration currently breaks that:

>>> import scipy, numpy as np
>>> assert scipy.array([1,2,3], dtype=scipy.float64).dtype == np.float64
__main__:1: DeprecationWarning: scipy.array is deprecated and will be removed in SciPy 2.0.0, use numpy.array instead
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/pauli/prj/scipy/scipy/scipy/_lib/deprecation.py", line 14, in call
    return fun(*args, **kwargs)
TypeError: data type not understood
>>> assert scipy.float64 == np.float64
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AssertionError

It's possible to do the decoration so that the above still works, via
subclassing with metaclass, but it's hard to be sure that covers everything
necessary. Since deprecation of functions already likely generates deprecation
warnings in suspect code, it's probably enough.

[rgommers]

Makes sense, thanks @pv.

Since this should go into 1.4.0rc1 and we're branching that today, I'll hit the green button. @larsoner would be good to still have a look

[larsoner]

Agreed that this is safer/better