DEP: drop mpmath as a depdendency, replaced with scipy

[neutrinoceros]

I think this is the correct replacement. However, I note that in my experiments with it, I hit some warnings in numpy, triggered from scipy:

/Users/clm/dev/gh/birnstiel/dsharp_opac/.venv/lib/python3.14/site-packages/scipy/_lib/array_api_compat/numpy/_aliases.py:106: ComplexWarning: Casting complex values to real discards the imaginary part

suggesting that scipy is maybe not solving correctly in the complex domain, or at the very least that it's doing something weird with numpy's API. I can probably dig it down further there, and I'll keep this PR as a draft until then.

[neutrinoceros]

update: the warning I saw earlier only appears when using scipy.optimize.root_scalar(..., method='newton'), not when calling scipy.optimize.newton directly. Thus, this is ready for review