cuda-core pinning is too tight

[leofang]

The documentation suggests that any version above 0.3.2 would work:

nvmath-python/docs/sphinx/installation.rst

Lines 334 to 339 in b92f2e6

	* - cuda-core
	-
	- >=0.3.2
	- >=0.3.2
	- >=0.3.2
	- >=0.3.2

But it is pinned too tightly in the dependency list:

nvmath-python/pyproject.toml

Line 37 in b92f2e6

"cuda-core >=0.3.2,<0.4",

and this causes problems for downstream users, developers, and our product managers to teach and use all Python projects together (example).

I suggest to loosen the upper bound to <1, following numba-cuda:
https://github.com/NVIDIA/numba-cuda/blob/358f3c4a38577d5bc61e5dbe228dddd3e8505d01/pyproject.toml#L29