👨‍🌾 Fix NaN values bound numpy windows version

[Crola1702]

🪰 BugFix

Issue

• The error is related to an unexpected keyword argument 'equal_nan' (Failing test in rosidl PR).
• We noticed this is a feature from numpy >= 1.19.0 (See ENH: Add equal_nan keyword argument to array_equal numpy/numpy#16128).
• Also, Windows Debug job has pinned numpy version to 1.18.4 here.

Reference build: Nightly Windows Debug

Log output:

>       assert numpy.array_equal(arr_of_float32_with_nan, msg.float32_values, equal_nan=True)

C:\ci\ws\src\ros2\rosidl_python\rosidl_generator_py\test\test_interfaces.py:508: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

args = (array([-1.33,   nan,  1.33], dtype=float32), array([-1.33,   nan,  1.33], dtype=float32))
kwargs = {'equal_nan': True}

>   ???
E   TypeError: _array_equal_dispatcher() got an unexpected keyword argument 'equal_nan'

<__array_function__ internals>:4: TypeError

Investigation

The proposed solution changes this test behavior to support both numpy <= 1.18 and >= 1.19.

Selected implementation: https://github.com/rossbar/numpy/blob/main/numpy/core/numeric.py#L2475

asarray(a1[~a1nan] == a2[~a1nan]).all() # plus shape verification

Windows build:
Linux aarch64 build:
Linux build:

[Blast545]

LGTM

[Blast545]

Do you think this approach is OK? @sloretz (tagging you as reviewer of the original PR)

[clalancette]

Seems reasonable enough to me.

The Rpr job is currently expected to fail here, so we can ignore that and just use the CI results.

[fujitatomoya]

@mergify backport humble

[mergify]

backport humble

❌ Command disallowed due to command restrictions in the Mergify configuration.

• sender-permission>=write

[fujitatomoya]

@Mergifyio backport humble

[mergify]

backport humble

❌ Command disallowed due to command restrictions in the Mergify configuration.

• sender-permission>=write

[fujitatomoya]

@clalancette @sloretz this one too, what do you think?