* 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.

One of this is a small issue exposed by new warnings, the others are
simply adapting our test suit to stricter integer coercion rules
(avoiding float -> int conversions).

The last one is that we assumed pickle protocol 5 would be in 3.8.
It is not yet included in the alpha releases at least.
It seems not necessary for the numpy test suit to check whether
it is available based on the python version so removing that test.
(Also testing if the pickle5 module works seems unnecessary.)

Closes gh-13412

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

Use standard pytest markers everywhere in the numpy tests. At this point
there should be no nose dependency. However, nose is required to test
the legacy decorators if so desired.

At this point, numpy test cannot be run in the way with runtests, rather
installed numpy can be tested with `pytest --pyargs numpy` as long as
that is not run from the repo. Run it from the tools directory or some
such.

The aim here is to separate out the nose dependent files prior to adding
pytest support. This could be done by adding new files to the general
numpy/testing directory, but I felt that it was to have the relevant
files separated out as it makes it easier to completely remove nose
dependencies when needed.

Many places were accessing submodules in numpy/testing directly, and in
some cases incorrectly. That presented a backwards compatibility
problem. The solution adapted here is to have "dummy" files whose
contents will depend on whether of not pytest is active. That way the
module looks the same as before from the outside.

In the case of numpy itself, direct accesses have been fixed. Having
proper `__all__` lists in the submodules helped in that.

Enforces that stacklevel is used in warnings.warn and "ignore"
is not used in filterwarnings or simplefilter for most of numpy.