Implement better testing and against multiple NumPy versions

• Testing with both NumPy v1.X and NumPy v2.X until NumPy v1 is obsolete
• #629 is going to add a schedule to the tests, extend the tests against NumPy and SciPy nightlies there (or in another workflow)
• Implement an upper bound on NumPy according to the NumPy guidelines
• Establish a 1.X lower-bound for our dependence on NumPy, i.e., find the minimum value of X where all tests pass and keep bumping it with every release

Notes to refer to, cross-posted from https://github.com/HIPS/autograd/pull/628#issuecomment-2305619693:

1. NumPy development docs
   • numpy.org/devdocs/dev/depending_on_numpy.html#testing-against-the-numpy-main-branch-or-pre-releases
   • numpy.org/devdocs/dev/depending_on_numpy.html#runtime-dependency-version-ranges
2. Package indices to grab nightly wheels from for NumPy and SciPy
   • anaconda.org/scientific-python-nightly-wheels/repo