Raise SymbolicUtils floor for minimum dependency tests

[ChrisRackauckas-Claude]

Ignore this PR until it has been reviewed by @ChrisRackauckas.

Summary

• raise the SymbolicUtils compatibility floor to 4.38.1
• assert that the HeatingDiode simulation terminates successfully before inspecting its solution

Root cause

The minimum-dependency workflow started selecting SymbolicUtils 4.10.0 after SymbolicUtils became a direct dependency. With the workflow's ModelingToolkit 11.2.0 / OrdinaryDiffEq 6.103.0 stack, SymbolicUtils 4.10.0 leaves a -3.48e-14 algebraic residual at the reverse-current bound. The computed minimum current is consequently -1.0000000347787363e-6, just below the model's existing physical assertion of -1e-6.

The same diagnostic stack produces exactly -1e-6 with SymbolicUtils 4.38.1 and 4.40.0. Tightening the solver tolerances did not remove the sign-sensitive residual with 4.10.0, so this keeps the test assertion unchanged and raises the dependency floor to the first verified good release instead.

Validation

• Runic 1.7.0 --check Project.toml test/analog.jl: passed
• exact Julia 1.10 minimum-dependency workflow with SymbolicUtils 4.38.1: Core/analog.jl | 100 pass, 100 total
• the full test command proceeded past analog and then encountered the unrelated clean-main IfElse test-dependency error in Core/chua_circuit.jl; that blocker is already addressed by draft PR test: use Base.ifelse in Chua circuit #477

Failing CI reproduced from https://github.com/SciML/ModelingToolkitStandardLibrary.jl/actions/runs/29193734380/job/86652873871.

[ChrisRackauckas-Claude]

CI checkpoint:

• Exact local downgraded Core/analog.jl passed 100/100 with SymbolicUtils 4.38.1.
• CI independently passes Core/analog.jl 100/100 on Julia current, LTS, prerelease, and the downgrade job.
• Each Core/downgrade job then stops at the unrelated clean-main Chua test dependency error, which is addressed by draft PR test: use Base.ifelse in Chua circuit #477.
• QA, Runic, suggestions, spelling, and matrix detection are green.
• The docs job reaches separate clean-main custom_component.md and input_component.md example failures; those are being reproduced independently without weakening Documenter error handling.

No assertion was loosened or skipped. The HeatingDiode successful-retcode regression and existing strict current bound both pass.

[ChrisRackauckas-Claude]

Post-merge downstream CI classification:

• ModelingToolkit.jl/All completed after 2h23m47s.
• Earlier MTK groups passed, including InterfaceII (945 pass / 7 existing broken) and SymbolicIndexingInterface (1964/1964).
• The terminal Downstream group errored only when its environment resolved LinearSolve 4.3.0 after the parent test process had already loaded LinearSolve 3.87.0. LinearSolveArnoldiMethodExt then tried to access LinearSolve.ArnoldiMethodJL on the already-loaded older module; the log explicitly reports LinearSolve v4.3.0 [loaded: v3.87.0].
• This is the exact mixed-version clean-master condition previously reproduced and fixed in closed ModelingToolkit PR #4730. Per the prior closure, I am not reopening or duplicating that PR.

This failure occurs after the SymbolicUtils-floor change's relevant MTK groups passed and is not caused by this PR.