Make uncertainties compatible with numpy 2.0

[eendebakpt]

In numpy 2.0 the numpy.core was renamed to numpy._core. Is this PR we change the imports in the uncertainties package to work with both numpy 2.0 and numpy 1.x.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 3 lines in your changes missing coverage. Please review.

Project coverage is 48.31%. Comparing base (35ab5e8) to head (4e0b451).

Files	Patch %	Lines
uncertainties/unumpy/core.py	0.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #245      +/-   ##
==========================================
- Coverage   48.34%   48.31%   -0.03%     
==========================================
  Files          12       12              
  Lines        1903     1902       -1     
==========================================
- Hits          920      919       -1     
  Misses        983      983              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[newville]

@eendebakpt Thanks! I agree.

Separately: I wonder if we should explicitly test with both numpy 1.2something and numpy 2.0.... but that should not be necessary here, as this probably does really warrant a bugfix release for numpy2 users.

[wshanks]

Looks good now!

[wshanks]

@newville We effectively do test on numpy 1 and 2 because we test on Python 3.8 and numpy 2.0 only supports Python 3.9+.

We may want to set deprecation warnings as test failures, or maybe just deal with the numpy.matrix pending deprecation warnings. I think I missed this deprecation warning about numpy.core amongst all the numpy.matrix pending deprecation warnings when I tested with numpy 2.0 before.

[newville]

@reneeotten @eendebakpt Thanks!

@wshanks good point about Py38 being pinned to Numpy1. I am okay with leaving this as it is.
I know of a few third-party packages that I use in downstream projects that have not yet released Numpy2-compatible code. I suspect it will take several months to safely assume that everything works with Numpy2. I'd hate for uncertainties (or lmfit) to either be the "stuck on Numpy1" bottleneck or require Numpy2 too early.

[reneeotten]

please squash the commits and update the commit message.

requests were satisfied.