[BUG] test failures in shared memory variant of 2D filtering kernels on Windows

[grlee77]

Describe the bug
For a Windows 11 system with CuPy 13.3.0, NumPy 1.26.4, CUDA 12.6 and an RTX A3000 GPU, some test failures were seen in some tests of separable filters:

Steps/Code to reproduce bug
run tests in the file cucim/skimage/filters/tests/test_separable_filtering.py on Windows

Expected behavior
All tests should pass

Environment details (please complete the following information):
Windows 11 system with CuPy 13.3.0, NumPy 1.26.4, CUDA 12.6 and an RTX A3000 GPU

installed devolpment branch wheel from #765

Additional context
Proposed temporarily disabling use of these kernels on Windows in #770

See attached error logs

separable_failures_win.txt

[jakirkham]

Thanks Greg! 🙏

Some of the errors reference Python format types that are undefined on Windows. Namely q and Q, which refer to long long and unsigned long long respectively. Think those types of inputs can simply be skipped on Windows

The other failures look like some kind of overflow issue (very large numbers when small ones were expected). So maybe there is some conversion or typing issue happening

Edit: That isn't to say all test failures look like this in that output, but those two issues stick out