test: check SISO initialization via ODEProblem

[ChrisRackauckas-Claude]

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

Summary

• exercise the SISO initialization assertions through SciMLBase.ODEProblem
• keep both exact checks (xd == 1.0 and u == 1.0) unchanged
• remove the test-only OrdinaryDiffEq import because this test no longer solves a standalone initialization problem

This is a fresh, focused extraction of the SISO test-path correction discussed in the stale omnibus #452. It does not include any of #452's unrelated changes.

Why

PR #469 changed this test from the earlier explicit generate_initializesystem/mtkcompile/NonlinearProblem flow to InitializationProblem(sys, 0.0), then indexed that standalone problem's solution with symbols from the original ODE system.

A standalone InitializationProblem represents the generated initialization system. Its symbolic container is not a contract that every original ODE unknown or observed alias remains indexable, and its intermediate unknown/default representation varies across compatible dependency graphs:

• current graph: xd initializes to 1.0, while indexing so.u errors because that alias is not present in the generated initialization system
• preserved minimum graph (ModelingToolkit 11.0.0, ModelingToolkitBase 1.1.0, Symbolics 7.4.1, SciMLBase 2.125.0): the standalone path returns xd == 0.0 and u == 1.0
• on that same minimum graph, the pre-Use InitializationProblem() in the SISO tests (rebase of #465) #469 explicit path returns xd == 1.0 and u == 1.0

The public ODEProblem construction is the operation whose contract the test actually wants: it runs initialization for the ODE and retains the original system as the symbolic provider. Both original assertions therefore remain exact and continue to test the intended initialized values.

Local validation

• Julia 1.10.11, current resolved graph, complete test/utils.jl: 8/8 passed
• Julia 1.12.6, current resolved graph, complete test/utils.jl: 8/8 passed
• Julia 1.13.0-rc1 with --compiled-modules=no, current graph, complete test/utils.jl: 8/8 passed (source mode was needed because the local stdlib JLL cache had incompatible precompile flags)
• preserved exact minimum graph, focused public ODEProblem path: xd == 1.0, u == 1.0
• whole-repository Runic --check .: exit 0
• git diff --check: exit 0

GROUP=Core julia +1.12 --project=. -e 'using Pkg; Pkg.test()' ran all preceding Core files successfully, then exited 1 at the known clean-main AccelerationSensor error in test/translational.jl:235 (InvalidSystemException, structurally singular mass₊v(t)). That unrelated base failure is addressed separately by #490; this PR does not alter or include that fix.

Investigation steps

1. Reproduced both reported SISO symptoms on an unmodified current checkout and a preserved exact minimum manifest.
2. Inspected the compiled ODE system and generated initialization systems to identify the symbol-container mismatch.
3. Re-ran the pre-Use InitializationProblem() in the SISO tests (rebase of #465) #469 path on the exact minimum graph to locate the source boundary.
4. Validated the public ODEProblem path on current Julia 1.10/1.12/1.13 and the exact minimum dependency graph.
5. Ran the repository-native Core harness, whole-repository Runic, and whitespace checks.

[ChrisRackauckas-Claude]

Closing this draft as a duplicate of #491, which already carries the focused SISO correction and its validation. No changes from this branch need to be merged separately.