This means that warnings of different origin then the one tested for
behave normally. If the normal behaviour is to igonre them this might
decrease specificity in rare cases. In most cases the specificity
will be slightly higher.

Also modify the corresponding test to suppress the non Deprecation
warnings created to test specificity.

This context has a couple of advantages over the typical one, and can
ensure that warnings are cleanly filtered without side effect for
later tests.

This changes the check for scalars in assert_almost_equal so that

    abs(actual - desired) < 1.5 * 10**(-decimal)

Note that the previous documentation claimed that the functions were
equivalent to

    abs(actual - desired) < .5 * 10**(-decimal)

but that was not how they behaved in practice.

Due to the change in implementation, some very delicate tests may fail
that did not fail before. No extra failures were noted in scipy.

Closes #5200.

    assert_equal(np.array([0, 1]), np.matrix([0, 1]))

used to print

     x: array([0, 1])
     y: [repr failed]

now prints

     x: array([0, 1])
     y: [repr failed for <matrix>: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()]

Revert "Merge pull request #7001 from shoyer/NaT-comparison"

This reverts commit 7141f40b58ed1e7071cde78ab7bc8ab37e9c5983,
reversing changes made to 8fa6e3bef26a6d4a2c92f2824129aa4409be2590.

The original broke some pandas tests. The current plan to get this
in is

* reversion
* issue FutureWarning in 1.11 and 1.12
* make the change in 1.13.