Raise SymbolicIndexingInterface compat lower bound to 0.3.47

[ChrisRackauckas-Claude]

Ignore until reviewed by @ChrisRackauckas.

What changed and why

Raises the SymbolicIndexingInterface compat lower bound to 0.3.47 in lib/ModelingToolkitBase/Project.toml (was 0.3.39) and the root Project.toml (was 0.3.46). Nothing else changes.

#5131 added the testsets "Array unknowns on a completed system" to lib/ModelingToolkitBase/test/odesystem.jl and test/sdesystem.jl. They index an array unknown on a completed (non-scalarized) system (prob[x], sol[x]). With SymbolicIndexingInterface ≤ 0.3.46, _getsym(sys, ::ArraySymbolic, ...) recurses into getsym(sys, idx::Vector{Int}), which calls is_observed(sys, ::Int) for every index. variable_symbols(csys) on a completed system is [x, w] (length 2), so index 3 is not "a variable", is_observed continues to parameter_index(sys, 3) → parameter_index(::IndexCache, ::Int), which has no method, and the SII fallback fails with MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache). SymbolicIndexingInterface 0.3.47 (SciML/SymbolicIndexingInterface.jl@373e18ee9, "Avoid recursive getsym call") returns GetStateIndex(idx) directly and never takes that path. Regular CI resolves 0.3.55 and passes; only the downgrade jobs, which resolve the floor (0.3.46), hit it.

Noticed on #5148, which does not touch these tests. It is a master failure:

• master Downgrade Sublibraries run for 444ccfb (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34779646406): downgrade-mtkbase (InterfaceII) (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34779646406/job/103820901951) errors at sdesystem.jl:1239,1249; downgrade-mtkbase (InterfaceI) (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34779646406/job/103820890316) errors at odesystem.jl:1788,1793,1803,1820, all with this MethodError.
• master root Downgrade / Downgrade Tests - InterfaceI for 444ccfb (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34779646422/job/103784179636): same four odesystem.jl errors.
• The previous master commit 8689e54 passed both workflows (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34778665578, https://github.com/SciML/ModelingToolkit.jl/actions/runs/34778665366).

Verification (Julia 1.10.11, local Pkg.develop of lib/ModelingToolkitBase with the downgrade job's versions pinned: SciMLBase 3.54.0, SymbolicUtils 4.37.0, Symbolics 7.37.0, DiffEqBase 7.18.1, StochasticDiffEq 7.1.3, OrdinaryDiffEq 7.8.1)

The scripts are the "Array unknowns on a completed system" testsets extracted verbatim from test/sdesystem.jl and test/odesystem.jl (plus a pkgversion print).

Before (master 444ccfb0, SymbolicIndexingInterface pinned to 0.3.46):

$ julia +1.10 --project=envdown sde_array_test.jl
SII=0.3.46 SciMLBase=3.54.0 Symbolics=7.37.0 SymbolicUtils=4.37.0 StochasticDiffEq=7.1.3
Array unknowns on a `complete`d system: Error During Test at sde_array_test.jl:26
  Expression: prob[x] ≈ [1.0, 2.0, 3.0]
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
    [1] parameter_index(indp::ModelingToolkitBase.IndexCache, sym::Int64)
    [2] parameter_index  @ lib/ModelingToolkitBase/src/systems/abstractsystem.jl:196 [inlined]
    [3] is_observed      @ lib/ModelingToolkitBase/src/systems/abstractsystem.jl:380 [inlined]
    ...
Array unknowns on a `complete`d system: Error During Test at sde_array_test.jl:36
  Expression: length((sol[x])[end]) == 3
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
ERROR: LoadError: Some tests did not pass: 8 passed, 0 failed, 2 errored, 0 broken.

$ julia +1.10 --project=envdown ode_array_test.jl
SII=0.3.46 SciMLBase=3.54.0 Symbolics=7.37.0 SymbolicUtils=4.37.0 DiffEqBase=7.18.1 OrdinaryDiffEq=7.8.1
  Expression: prob[x] ≈ [1.0, 2.0, 3.0]
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
  Expression: ≈((sol[x])[end], (analytic(1.0))[1:3], rtol = 1.0e-6)
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
  Expression: ≈((msol[x])[end], (sol[x])[end], rtol = 1.0e-6)
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
  Expression: ≈((dsol[x])[end], (analytic(1.0))[1:3], rtol = 1.0e-5)
  MethodError: no method matching symbolic_container(::ModelingToolkitBase.IndexCache)
ERROR: LoadError: Some tests did not pass: 20 passed, 0 failed, 4 errored, 0 broken.

After (this branch; Pkg.free + Pkg.add(name="SymbolicIndexingInterface", version="0.3.47") + Pkg.pin, everything else unchanged):

$ julia +1.10 --project=envafter sde_array_test.jl
SII=0.3.47 SciMLBase=3.54.0 Symbolics=7.37.0 SymbolicUtils=4.37.0 StochasticDiffEq=7.1.3
Test Summary:                          | Pass  Total   Time
Array unknowns on a `complete`d system |   10     10  18.5s

$ julia +1.10 --project=envafter ode_array_test.jl
SII=0.3.47 SciMLBase=3.54.0 Symbolics=7.37.0 SymbolicUtils=4.37.0 DiffEqBase=7.18.1 OrdinaryDiffEq=7.8.1
Test Summary:                          | Pass  Total   Time
Array unknowns on a `complete`d system |   24     24  27.4s

The whole lib/ModelingToolkitBase/test/sdesystem.jl also passes in the same after-environment (included directly; 16 top-level testsets, no failures, no errors):

$ julia +1.10 --project=envafter run_sdesystem_file.jl
SII=0.3.47
...
Test Summary:                          | Pass  Total  Time
Array unknowns on a `complete`d system |   10     10  6.3s
SDESYSTEM_FILE_DONE

typos and Runic report nothing on the two changed files.

Not verified

• I did not run the full Downgrade / Downgrade Sublibraries workflows locally, only the affected testsets and sdesystem.jl with the job's pinned versions.
• Not the whole odesystem.jl file with SII 0.3.47 (only its "Array unknowns on a completed system" testset, 24/24).
• The master Downgrade Sublibraries (InterfaceI) job for 444ccfb also has two unrelated errors in test/array_equation_dae.jl:60,118 (a parameter-despecialization barrier requires exactly one parameter wrapper, thrown in DiffEqBase 7.18.1 solve.jl:842). They are not touched by this PR; that file is unchanged since 2026-08-17 and the previous green master run resolved DiffEqBase 7.15.0, so it looks like a DiffEqBase-floor issue that surfaced when the resolver moved to 7.18.1.
• With ModelingToolkit (tearing) loaded, the raw mprob.u0 ≈ prob.u0 / mprob.f(...) ≈ prob.f(...) comparisons in the same testsets fail on the regular tests / InterfaceI and InterfaceII jobs for 444ccfb (https://github.com/SciML/ModelingToolkit.jl/actions/runs/34779646543). That is a separate regression from Support array unknowns on a completed system for every problem type #5131 (unknown order after mtkcompile), handled in Compare complete/mtkcompile array-unknown problems through symbolic indices #5154.

🤖 Generated with Claude Code (model: claude-fable-5-1)
https://claude.ai/code/session_01VorhgJYhDasyVLpGNrviga