Bump downstream OrdinaryDiffEq/MTK compat for RAT 4.x

[ChrisRackauckas-Claude]

Summary

Restores Tests - Downstream to green. Master has been red on Downstream Julia 1/lts/pre for weeks; this PR makes the resolver succeed and updates the downstream sources for the OrdinaryDiffEq 7 API the bump exposes.

Compat changes (test/downstream/Project.toml)

• OrdinaryDiffEq = "6.31" → "6.31, 7"
• ModelingToolkit = "8.33, 9" → "8.33, 9, 10, 11" (OrdinaryDiffEq 7 transitively wants MTK 10+)
• Add OrdinaryDiffEqRosenbrock = "1, 2" (no longer re-exported from OrdinaryDiffEq)
• Add RecursiveArrayToolsShorthandConstructors (used by the AP[...] shorthand introduced in 334bd95)

Why the resolver was unsatisfiable on master

• Adapt = "4" (required by RAT 4.x) narrows OrdinaryDiffEq to [4.0.0 - 5.39.1, 6.64.0 - 7.0.0].
• OrdinaryDiffEq 6.110+ pins RecursiveArrayTools = "3.52.0 - 3".
• Intersection in the 6.x window is empty, so OrdinaryDiffEq 7 is the only viable path.

Downstream source fixes (exposed once the resolver succeeds)

• Rosenbrock23(autodiff = false) → Rosenbrock23(autodiff = AutoFiniteDiff()). OrdinaryDiffEq 7 rejects Bool and asks for an ADType.
• Add using OrdinaryDiffEqRosenbrock to odesolve.jl (provides Rodas5).
• Add using RecursiveArrayToolsShorthandConstructors to odesolve.jl and downstream_events.jl (needed for AP[...]).
• SciMLBase.successful_retcode(sol) → unqualified successful_retcode(sol). SciMLBase is not a bare binding via using OrdinaryDiffEq, but successful_retcode and AutoFiniteDiff are — so we avoid adding SciMLBase/ADTypes as direct deps, which would unresolvably conflict with RAT 4.3's transitive SciMLBase 3 requirement.

After (this branch)

Test Summary: | Pass  Broken  Total     Time
ODE Solve     |    4       1      5  1m34.7s
Test Summary: | Total  Time
Events        |     0  2.4s
Test Summary:          | Broken  Total  Time
Measurements and Units |      1      1  4.2s
Test Summary: | Pass  Total  Time
TrackerExt    |    1      1  0.5s

Out of scope

• test/downstream/symbol_indexing.jl (run in the SymbolicIndexingInterface group) still has many failures, almost certainly from MTK 11 / SymbolicIndexingInterface API changes around symbolic-array indexing. Worth its own investigation — not gating this PR.
• test/downstream/adjoints.jl remains commented out in runtests.jl per the pre-existing # TODO: re-enable after SciMLBase compat bump for RAT v4 note.

Test plan

• Resolution succeeds on Julia 1.11 (Pkg.activate + Pkg.develop + Pkg.instantiate for test/downstream)
• ODE Solve Tests safetestset passes locally (4 pass, 1 pre-existing broken)
• Event Tests with ArrayPartition safetestset runs without error
• Measurements and Units safetestset passes (1 pre-existing broken)
• TrackerExt safetestset passes (1 pass)
• CI on SciML/RecursiveArrayTools.jl Tests - Downstream - Julia 1/lts/pre