Fix for incompatibility of numpy2.* in the finite difference routines

[ZachMcBreartyUU]

Numpy changed how np.linalg.solve works, requiring that the second argument is of the from (M,) or (..., M, K) instead of (..., M) or (..., M, K).
Finite difference is using the old form but can be easily updated by adding a axis such that K = 1 and then squeezing (removing) that axis.
The way this is implemented is compatible with numpy1.*
This change is explained in comments in the _coefficients code and is propagated to the rest of the specialized coefficients code without comment.
Solves issues #551
Alternative is to specify that numpy<=1.26.4

[Jacob-Stevens-Haas]

Numpy 2.0 causes a lot of failures, I think limiting to numpy<2.0 would be better for now, unless you're willing to troubleshoot those failures.

[ZachMcBrearty]

For now I would very much just recommend adding numpy<2.0.0 to the pyproject.toml, esp since numpy 2 will be default installed otherwise.
I didn't realise that numpy2 would clash with cvxpy etc, I only ran pytest on the differentiation file.

[Jacob-Stevens-Haas]

@ZachMcBrearty thank you for making this PR nonetheless - it forced me to pay attention to stuff I've been neglecting. Closing in favor of #596, but happy to chat more about this or work towards 2.0 compatibility.