test: prevent CI 6-hour timeout and fix Thermal/Magnetic regressions

This addresses the real root cause of CI hitting the 6-hour job
timeout: the `Thermal/thermal.jl` "FixedHeatFlow with alpha=0.0 test"
testset calls `solve(prob)` (no algorithm) on a compiled system whose
mtkcompile-reduced form has zero differential states. The default
auto-switching algorithm cannot make forward progress on that and
spins indefinitely instead of erroring or terminating. With an
explicit `ImplicitEuler(; dt=1.0, adaptive=false)`, the test finishes
in ~16s. (The retcode assertion still fails for this stateless system
- same as on April 27 - but the rest of the suite can now run to
completion.)

Other test fixes in the same area, found while reproducing locally:

* `Thermal/motor.jl`, `Thermal/piston.jl` - missing `using
  OrdinaryDiffEq` and `t_nounits as t` imports, causing
  `@mtkmodel`-expanded `iv` to be MTKBase's
  `PleaseImportDynamicQuantities()` sentinel and `solve(prob)` to
  error out at `Default algorithm choices require
  DifferentialEquations.jl`.
* `Magnetic/magnetic.jl` Inductor: the flux-conservation assertion at
  the shared magnetic node used exact `.==` against an observable
  reconstructed via symbolic substitution; the structural equation
  is enforced but rounding in the substitution accumulates to ~2e-19.
  Loosened to `isapprox(...; atol=1e-12)` - still 7 orders of
  magnitude below any physical flux scale.

Verified locally on MTK v11.26.0 / MTKBase v1.35.1 / Symbolics v7.23.0
/ SymbolicUtils v4.29.1:
- Thermal/thermal.jl: 18/19 pass (1 pre-existing fail in the
  alpha=0 testset, also failing on April 27 master)
- Thermal/motor.jl: 6/6
- Thermal/piston.jl: 3/3
- Magnetic/magnetic.jl: 3/3

Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>

Author: ChrisRackauckas

test/Magnetic/magnetic.jl

@@ -63,5 +63,15 @@ using OrdinaryDiffEqRosenbrock: Rodas4
 
     @test SciMLBase.successful_retcode(sol)
     @test sol[r_mFe.Phi] == sol[r_mAirPar.Phi]
-    @test all(sol[coil.port_p.Phi] + sol[r_mLeak.Phi] + sol[r_mAirPar.Phi] .== 0)
+    # Conservation of flux at the shared node is structural (a `connect` flow
+    # equation), but observed via numerical reconstruction. Roundoff in the
+    # observable substitution accumulates to ~1e-19; use isapprox with a
+    # tolerance well above that (and far below any physical scale).
+    @test all(
+        isapprox.(
+            sol[coil.port_p.Phi] + sol[r_mLeak.Phi] + sol[r_mAirPar.Phi],
+            0;
+            atol = 1.0e-12
+        )
+    )
 end

test/Thermal/motor.jl

@@ -1,4 +1,5 @@
-using ModelingToolkit, Test
+using ModelingToolkit, OrdinaryDiffEq, Test
+using ModelingToolkit: t_nounits as t, D_nounits as D
 using SciCompDSL
 using SciMLBase
 using ModelingToolkitStandardLibrary.Thermal

test/Thermal/piston.jl

@@ -1,4 +1,5 @@
-using ModelingToolkit, Test
+using ModelingToolkit, OrdinaryDiffEq, Test
+using ModelingToolkit: t_nounits as t, D_nounits as D
 using SciCompDSL
 using SciMLBase
 using ModelingToolkitStandardLibrary.Thermal

test/Thermal/thermal.jl

@@ -5,6 +5,7 @@ using ModelingToolkit: t_nounits as t, D_nounits as D
 using ModelingToolkitStandardLibrary.Blocks: Constant, Step
 using OrdinaryDiffEq: ReturnCode.Success
 using OrdinaryDiffEqRosenbrock: Rodas4
+using OrdinaryDiffEqSDIRK: ImplicitEuler
 
 # Test HeatCapacitor, TemperatureSensor, RelativeTemperatureSensor, FixedTemperature
 @testset "Heat systems" begin
@@ -195,7 +196,11 @@ end
     @info "Building a FixedHeatFlow with alpha=0.0"
     @mtkcompile test_model = TestModel() allow_parameter = false
     prob = ODEProblem(test_model, [test_model.wall.Q_flow => nothing, test_model.wall.dT => nothing], (0, 10.0); guesses = [test_model.heatflow.port.T => 1.0])
-    sol = solve(prob)
+    # The compiled system has no differential states (purely algebraic). The
+    # auto-switching default algorithm cannot make forward progress on this
+    # and hangs the test runner indefinitely; ImplicitEuler with a fixed dt
+    # handles the trivially-constant trajectory.
+    sol = solve(prob, ImplicitEuler(); dt = 1.0, adaptive = false)
 
     heat_flow = sol[test_model.heatflow.port.Q_flow]