Remove type instabilities and allow general floating point types

[JoshuaLampert]

There were several type instabilities. I tried to fight them as much as I could, but in some places it might be unavoidable (e.g., creating NodeSets from non-statically sized vectors). I also added some tests that the floating point type is preserved during interpolate and solve_stationary. It should cover most of the package, but maybe not everything.
I also decided to force floating point element types for NodeSets, which makes lives easier.

[JoshuaLampert]

This is breaking because:

1. random_hypersphere and random_hypersphere_boundary expect a Tuple as center now.
2. The element type of NodeSets will always be converted to a floating point type as this makes more sense in an interpolation framework. A similar approach is also taken by the Meshes.jl/CoordRefSystem.jl framework.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.66%. Comparing base (7002677) to head (4823e8d).
Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #121   +/-   ##
=======================================
  Coverage   98.65%   98.66%           
=======================================
  Files          46       46           
  Lines        1266     1269    +3     
=======================================
+ Hits         1249     1252    +3     
  Misses         17       17           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JoshuaLampert]

Another instance of OrdinaryDiffEq.jl completely breaking Downgrade... I think that's because of SciML/OrdinaryDiffEq.jl#2567. The latest version of OrdinaryDiffEqDifferentiation.jl (v1.6.0) (which is installed because we don't require a compat of it) is not compatible with older versions of OrdinaryDiffEqNonlinearSolve.jl and OrdinaryDiffEqRosenbrock.jl and therefore we need the latest ones (v1.6 and v1.9 respectively).

[JoshuaLampert]

Because of yet another wrongly set compat bound in the SciML ecosystem, we also need to drop support for DiffEqCallbacks.jl v3...