[ChrisRackauckas-Claude]

Summary

GROUP=ModelingToolkit is declared in test/test_groups.toml and CI schedules three lanes for it, but test/runtests.jl has no ModelingToolkit body: SciMLTesting's run_tests treats the unknown group as a fall-through to the root core = () -> nothing no-op (confirmed in SciMLTesting 2.13 source, master CI job logs where the test body finishes in ~1.3s after ~14 minutes of install/precompile, and a local SciMLTesting probe). This PR removes that empty [ModelingToolkit] entry so CI stops burning those lanes, and drops the stale Dependabot directory for the deleted test/modelingtoolkit Project.toml.

It also deletes orphaned root copies under test/modelingtoolkit/, test/multirate/, and test/multithreading/ that are never included from the root runner. Same-named tests already run in OrdinaryDiffEqNonlinearSolve, OrdinaryDiffEqDifferentiation, OrdinaryDiffEqMultirate, and OrdinaryDiffEqExtrapolation. The root copies never ran in any root lane, so their diffs vs the lib copies are not live coverage loss. Precise differences:

• jacobian_tests: root used Rodas5; lib uses Rodas5P.
• nlstep_tests: root had two ≈ atol=1e-3 checks where lib uses 5e-3, and root ran TRBDF2 with default smooth_est where lib deliberately uses smooth_est=false (Align NLStep TRBDF2 parity test estimator #4077). Lib is otherwise stricter: two @test_broken → @test, and the FBDF order tolerance goes from 0.3/0.35 to 0.05.
• mreef_tests: root compared ratios to 2^order within 15%; lib uses test_convergence atol=0.3 and adds exact nf/nf2 counts and a dense-output testset.
• preconditioners / dae_initialize_integration / ode_extrapolation_tests: differ only in imports and (for preconditioners) manual sparsity instead of MTK.

Verification

• SciMLTesting unknown-group behavior (local probe, SciMLTesting 2.13):

  GROUP=ModelingToolkit … detect_sublibrary_group => ("ModelingToolkit", "Core")
  isdir(lib/ModelingToolkit) => false
  root would route to sublib Pkg.test? false
  core body invoked
  fell through to core? true
  PASS: GROUP=ModelingToolkit is unknown -> empty core no-op path
  

• Master CI (gh run view 36096837419 --job 107957121978): after GROUP=ModelingToolkit and full precompile (~822s), the test phase is 1.257923 seconds then OrdinaryDiffEq tests passed with no safetestset/include of any modelingtoolkit file.
• Diffs + git log --follow for each deleted root file vs its sublibrary counterpart (precise diffs above).
• typos clean on touched files. No .jl content edited (only deletions / Dependabot YAML), so Runic not applicable.

Not verified

• Full local GROUP=ModelingToolkit julia --project -e 'using Pkg; Pkg.test()' was not re-run end-to-end (would only reconfirm the empty core after a long resolve/precompile); dispatch path and master CI log were used instead.
• Did not re-run the sublibrary ModelingToolkit / Multithreading / Multirate suites; they already run those files on master.

Please ignore this draft until reviewed by @ChrisRackauckas.

Risk assessment

• Risk: Low. Deletes unused root files and removes a CI matrix entry that runs no tests.
• Blast radius: Root CI only — drops three ModelingToolkit jobs; sublibrary test coverage unchanged. Also removes a Dependabot directory that no longer exists.
• Evidence: SciMLTesting fall-through docs/source; local dispatch probe; master CI log (~1.3s empty body); git history/diffs showing root copies never ran and differ from live sublib tests as listed above.
• Independent review: Claude Code (Opus 5.5 subagent) rated it low, high confidence: none of the deleted files ran in any root lane; each has a live sublibrary copy run by its CI (NonlinearSolve/Differentiation ModelingToolkit groups, Multirate Core, Extrapolation Multithreading); root GROUP=ModelingToolkit fell through to the no-op core; no SciML repo calls GROUP=ModelingToolkit on OrdinaryDiffEq. VERDICT MERGE.
• Merge: needs review

Harness: Cursor Agent CLI 2026.09.26-dd393fe; model: auto; transcript: /home/crackauc/sandbox/goals/qa-hygiene/jobs/ode-layout/log.txt on amdci2