Fix reinit! to propagate Initial(...) params with SciMLBase 3.6

[ChrisRackauckas-Claude]

Summary

Fixes 4 pre-existing master failures in OrdinaryDiffEqNonlinearSolve_ModelingToolkit CI (e.g. https://github.com/SciML/OrdinaryDiffEq.jl/actions/runs/25053979241/job/73389128637) — the \reinit!` updates initial parameterstestset inlib/OrdinaryDiffEqNonlinearSolve/test/modelingtoolkit/dae_initialize_integration.jl` (lines 98-101).

Root cause

SciMLBase 3.6.0 added a LateBindingUpdateU0PContext parameter to late_binding_update_u0_p. Its 6-arg wrapper now forwards through a new 8-arg (prob, root_indp, u0, p, t0, newu0, newp, ctx) overload, which dispatches to the SciMLBase identity stub. MTK's existing 7-arg overload late_binding_update_u0_p(prob, sys::AbstractSystem, u0, p, t0, newu0, newp) (without ctx) is no longer reached from reinit!. As a result, reinit!(integrator, new_u0) stops propagating the new state into Initial(x) / Initial(y) parameters: both integ.ps[Initial(x)] / integ.ps[Initial(y)] and integ[x] / integ[y] stay at their pre-reinit! values.

This is dispatch-driven and does not show up at all under SciMLBase < 3.6.

Fix

In OrdinaryDiffEqCore's reinit!, sidestep the new context-aware dispatch by computing root_indp = SciMLBase.get_root_indp(integrator.sol.prob) ourselves and calling the 7-arg form late_binding_update_u0_p(prob, root_indp, u0, ..., newp) (no ctx). That form still dispatches to MTK's 7-arg overload, which is the path that actually updates the Initial(...) parameters in MTKParameters.initials.

This keeps the change scoped to ODE Core (one-line plus comment) and works whether MTK upgrades to overload the new 8-arg form or not.

Test plan

• Reproduce the failure locally with ~/.juliaup/bin/julia +1.11 --project=lib/OrdinaryDiffEqNonlinearSolve/test/modelingtoolkit lib/OrdinaryDiffEqNonlinearSolve/test/modelingtoolkit/dae_initialize_integration.jl against master — 4 fails on reinit! testset
• Apply fix — DAE Initialize Integration: 21/21 passes locally; reinit! updates initial parameters: 8/8
• Re-ran NLStep Tests (19 pass / 6 broken — unchanged), Preconditioner Tests (18/18) locally to confirm no regression in adjacent groups
• Runic check passes on lib/OrdinaryDiffEqCore/src/integrators/integrator_interface.jl
• Wait for SublibraryCI OrdinaryDiffEqNonlinearSolve_ModelingToolkit to confirm green