- Always enable __array_function__ overrides.
- Remove special cases for Python 2 compatibility.
- Document these changes in 1.17.0-notes.rst.

It will be good to see ASV numbers to understand the performance implications
of these changes. If need be, we can speed up NumPy functions internally by
using non-dispatched functions (with ``.__wrapped__``).

Example behavior:

>>> x = np.array([1, 2, 3])
>>> y = np.array([1, 2, 3.0001])
>>> np.testing.assert_allclose(x, y)
AssertionError:
Not equal to tolerance rtol=1e-07, atol=0

Mismatch: 33.3%
Max absolute difference: 0.0001
Max relative difference: 3.33322223e-05
 x: array([1, 2, 3])
 y: array([1.    , 2.    , 3.0001])

Motivation: when writing numerical algorithms, I frequently find myself
experimenting to pick the right value of `atol` and `rtol` for
`np.testing.assert_allclose()`. If I make the tolerance too generous, I risk
missing regressions in accuracy, so I usually try to pick the smallest values
for which tests pass. This change immediately reveals appropriate values to
use for these parameters, so I don't need to guess and check.

* ported the refguide_check module from SciPy for usage
in NumPy docstring execution/ verification; added the
refguide_check run to Azure Mac OS CI

* adjusted NumPy docstrings such that refguide_check passes

* Review F401,F841,F842 flake8 errors (unused variables, imports)

* Review comments

* More tests in test_installed_npymath_ini

* Review comments

BUG: Fix misleading assert message in assert_almost_equal #12200

Fixes #12200 by making a copy of the matrix before NaN's are excluded.
Add a test for it.

* ENH: __array_function__ support for np.lib, part 2

xref GH12028

np.lib.npyio through np.lib.ufunclike

* Fix failures in numpy/core/tests/test_overrides.py

* CLN: handle depreaction in dispatchers for np.lib.ufunclike

* CLN: fewer dispatchers in lib.twodim_base

* CLN: fewer dispatchers in lib.shape_base

* CLN: more dispatcher consolidation

* BUG: fix test failure

* Use all method instead of function in assert_equal

* DOC: indicate n is array_like in scimath.logn

* MAINT: updates per review

* MAINT: more conservative changes in assert_array_equal

* MAINT: add back in comment

* MAINT: casting tweaks in assert_array_equal

* MAINT: fixes and tests for assert_array_equal on subclasses

Otherwise the deprecation message is not that helpful as I have to dig
through the source to find out since when and whether conditional import
is necessary in my code. I may not have numpy on my dev machine (this
message was on CI). So it's one extra step for the consumer.

That's also for you, you have more chance of having consumer update
their code when they have more informations in the deprecation message,
so you can drop old code faster !

At least kudo to Charles Harris for including a comments just above,
that prevented me from having to git blame and go spelunking for knowing
which versions were or not affected.