fix: correct dtypes for numpy v2

[ianna]

fixes #3160

[ianna]

@jpivarski - this PR fixes the issue with Numpy v2: typically, np.intp is used to represent the platform integer type and the check for fallback on it is valid only for Numpy version 2 and higher.

[jpivarski]

Great! Thanks for finding and fixing this. However, it's not fully testing the platform-version combinations users might have. The

awkward/.github/workflows/test.yml

Lines 52 to 54 in 02858f8

	- python-version: '3.9'
	python-architecture: x86
	runs-on: windows-latest

test used to check 32-bit Windows with NumPy < 2, now it tests 32-bit Windows with NumPy ≥ 2. However, nothing tests 32-bit Windows with NumPy < 2 anymore, and we need to be sure that the _use32 evaluates to the right boolean value for all of those cases. This test should be split into two tests; one for NumPy < 2, one for NumPy ≥ 2, and both for 32-bit Windows.

While adding that 1 new test, these 3 old tests can be removed:

awkward/.github/workflows/test.yml

Lines 64 to 75 in 02858f8

	- python-version: '3.11'
	python-architecture: x64
	runs-on: ubuntu-latest
	dependencies-kind: numpy2
	- python-version: '3.11'
	python-architecture: x64
	runs-on: macos-11
	dependencies-kind: numpy2
	- python-version: '3.11'
	python-architecture: x64
	runs-on: windows-latest
	dependencies-kind: numpy2

which implies that we can also delete this file:

https://github.com/scikit-hep/awkward/blob/main/requirements-test-numpy2.txt

When those tests are changed, I'll have to update the set of required tests, but just request a review again. (These are all adjusting for the fact that NumPy has been officially released.)

[ianna]

@jpivarski - please, update required tests. Thanks!

[jpivarski]

At first, I thought that "Tests / Run Tests (pypy3.9, x64, ubuntu-latest, pypy)" was another new test, since it's not required yet, either. But now I remember that it's there so we can keep an eye on it, not so that we can promise that Awkward works on PyPy. (We don't provide a wheel for that.)

So, this gets us entirely up-to-date with NumPy 2.0 in its fully released state. This is ready to merge into main, and it can fix #3136 when that is subsequently synced with main.