Fix isothermal domain test connections

[ChrisRackauckas-Claude]

This PR should be ignored until reviewed by @ChrisRackauckas.

Summary

• Use domain_connect for hydraulic domain propagation instead of routing flow equations through the fluid parameter setter.
• Keep the cap/volume physical connection explicit.
• Transfer the Actuator operating point with public symbolic problem indexing instead of pairing symbolic unknowns with raw solution storage.
• Check domain parameter propagation through the public ModelingToolkitBase.bindings API.
• Add source/volume flow-balance assertions to the DynamicVolume test.

Root cause

ModelingToolkit commit b0f315369 stopped silently zeroing underconstrained variables during alias elimination. The immediate parent 80f7f62bc passes the reduced DynamicVolume model, while b0f315369 reports 48 highest-order variables and 47 equations. The tests were using a fluid domain setter as a physical flow connector, which hid the model structure behind the removed zeroing behavior.

The Actuator test also copied a failed zero-length solution by zipping unknowns(initsys) with raw initsol.u storage. Its public problem operating point contains piston.dx = 0, while the copied raw value was NaN. Reading each unknown through initprob[v] preserves the intended operating point.

Validation

• Rebased head 1cb83378 on current main 8539d8f2 (merged test: use Base.ifelse in Chua circuit #477).
• Direct test/isothermal_compressible.jl on Julia 1.12.6, exit 0:
  • Fluid Domain and Tube: 2/2
  • Valve: 1/1
  • DynamicVolume and minimum-volume feature: 12/12
  • Actuator System: 9/9
  • Prevent Negative Pressure: 3/3
• Exact commit boundary with the same reduced model and local environment:
  • 80f7f62bc: exit 0
  • b0f315369: exit 1, 48 variables / 47 equations
• Official GROUP=Core Pkg.test() on a temporary validation stack containing current main, this patch, and Fix callable interpolation builder symtype #480: Analog 100/100, Chua 1/1, Continuous 76/76, Digital 34/34, and Isothermal 27/27; the run then stopped on the separate clean-main exact-equality assertion in test/magnetic.jl:66, which is under an independent reproduction and ULP analysis.
• Repository-wide Runic 1.7 --check .: exit 0 across 81 files.
• git diff --check: exit 0.

No test was skipped, disabled, silenced, or made warn-only.

Related: #457, merged #477, and #480.

[ChrisRackauckas-Claude]

Rebase and validation update:

• Upstream main advanced to 8539d8f2 when test: use Base.ifelse in Chua circuit #477 merged. I stopped before pushing, fetched both remotes, and verified the feature remote still contained only the expected old head 7b7966a9.
• Rebased the single isothermal commit onto that new main; new head is 1cb83378 and the required co-author trailer is preserved.
• Re-ran the complete focused file on Julia 1.12.6: 27/27 assertions passed with isothermal_rebase_test_exit=0.
• Re-ran Runic across all 81 repository files and git diff --check: both exited 0.
• Updated the remote with an exact-SHA --force-with-lease; CI has restarted.

The official integration Core run now reaches and passes Isothermal 27/27, then exposes only the separately audited magnetic one-ULP assertion.

[ChrisRackauckas-Claude]

Post-rebase native GROUP=Core Pkg.test() evidence on Julia 1.12.6 (head 1cb83378):

• Core/analog.jl: 100/100
• Core/chua_circuit.jl: 1/1
• Core/continuous.jl: 76/76
• Core/demo.jl: 1/1
• Core/digital.jl: 34/34
• Core/isothermal_compressible.jl: 27/27

The group then stopped at Core/magnetic.jl (2 pass, 1 exact-equality failure). That same one-ULP-scale flux residual reproduces on unmodified current main and is being handled in a separate clean-main audit; it is not introduced by this PR.