All errors raised within the sub process will remain silenced if the return code isn't validated

This adds a warning when the main NumPy module is reloaded
with the assumption that in this case objects such as `np.matrix`,
`np._NoValue` or exceptions may be cached internally.

It also gives a warning when NumPy is imported in a sub-interpreter.

This is technically not a bug, but some IDEs and IPython have
autoreload magic which can mean that NumPy gets reloaded a second
time. This is not safe, but when it happens ignoring that an
identical docstring is already attached fixes the issue.

* Cleanup unused imports (F401) of mostly standard Python modules,
  or some internal but unlikely referenced modules
* Where internal imports are potentially used, mark with noqa
* Avoid redefinition of imports (F811)

As numpy is Python 3 only, these import statements are now unnecessary
and don't alter runtime behavior.

The pickle module was being imported from numpy.core.numeric. It was
defined there in order to use pickle5 when available in Python3 and
cpickle in Python2. The numpy.compat module seems a better place for
that.

All imports of pickle from numpy modules are now done this way:
>>> from numpy.core.numeric import pickle

Also, some loops on protocol numbers are added over pickle tests that
were not caught from #12090

That function is nose specific and has not worked since `__init__` files
were added to the tests directories.