Fixes the previously failing test cases on my Win10, debug build, 64-bit target, no AVX2 cpu.

[==========] Running 6 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 6 tests from Core/CountNonZeroBig
[ RUN      ] Core/CountNonZeroBig./0, where GetParam() = (0, 1x524190)
[       OK ] Core/CountNonZeroBig./0 (4 ms)
[ RUN      ] Core/CountNonZeroBig./1, where GetParam() = (0, 524190x1) 
[       OK ] Core/CountNonZeroBig./1 (5 ms)
[ RUN      ] Core/CountNonZeroBig./2, where GetParam() = (0, 3840x2160)
[       OK ] Core/CountNonZeroBig./2 (10 ms)
[ RUN      ] Core/CountNonZeroBig./3, where GetParam() = (5, 1x524190)
[       OK ] Core/CountNonZeroBig./3 (8 ms)
[ RUN      ] Core/CountNonZeroBig./4, where GetParam() = (5, 524190x1)
[       OK ] Core/CountNonZeroBig./4 (27 ms)
[ RUN      ] Core/CountNonZeroBig./5, where GetParam() = (5, 3840x2160)
[       OK ] Core/CountNonZeroBig./5 (53 ms)
[----------] 6 tests from Core/CountNonZeroBig (128 ms total)

For clarity, I want to raise a caution on what you wrote above "(MacOSX and Win32 platforms doesn't have AVX2 IPP optimizations)". It is my understanding that some OpenCV builds on Win32 platforms do support AVX2 IPP optimizations. However, 32-bit compile targets do not. When writing, I believe it is important to distinguish between the "Win32 API" and a Windows 32-bit target so that readers don't confuse the two. :-)

Right, Win32 above means "Windows 32-bit" (not Win32 API) in terms of OpenCV CI builds (64-bit is "Win64").

Also this note is correct for current IPPICV package (subset of Intel IPP functions and optimizations) only.
Standalone (external) IPP package or future updates of IPPICV may support AVX2 too (at least on Mac).

Is the cherry-pick into master/4.5 branch automated? I don't see this change yet there.
Just checking as an additional set of eyes 🤪

It is not automated, but it is regular (weekly / bi-weekly): https://github.com/opencv/opencv/wiki/Branches

You may want to pick this related patch too: #18991