Remove `OrdinaryDiffEq` internals

[SouthEndMusic]

Fixes https://github.com/Deltares/Ribasim/issues/1778.

See also https://github.com/SciML/OrdinaryDiffEq.jl/pull/2443.

[SouthEndMusic]

Things I changed:

• Replace LineSearches.jl by LineSearch.jl (part of SciML). Specifically, I use LineSearch.BackTracking instead of my wrapper of LineSearches.BackTracking. It might turn out that this wrapper is still needed.
• Use the latest master of OrdinaryDiffEq.jl (I had quite some problems installing directly from the GitHub repository, so I now use a path to the repo locally as a POC). Using NonlinearSolveAlg now works with finite difference, but using ForwardDiff is still broken. OrdinaryDiffEqRosenbrock is also broken.

It must be noted that even though NonlinearSolveAlg with finite difference and BackTracking from LineSearch.jl works, I observed a huge drop in performance for the HWS model. Interestingly, looking at the profile reveals that this performance drop is at least in part due to runtime dispatch and allocations in OrdinaryDiffEqNonlinearSolve, so that subpackage is clearly not mature enough to be used.