test: restore working MTKBase BVP assertions

[ChrisRackauckas-Claude]

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

Summary

• restore the four ModelingToolkitBase BVP solve paths that currently succeed
• assert their solution values and successful return codes instead of passing solution objects to Test.@test_broken
• leave the genuinely broken parameter-estimation cases marked broken

Cause

Commit f71cc7d wrapped four successful BVP solves in Test.@test_broken. A successful solve returns a solution object, not a Boolean, so Test reports each case as an error rather than a broken test.

This is the same failure in BoundaryValueDiffEq clean master run https://github.com/SciML/BoundaryValueDiffEq.jl/actions/runs/30342212061 and PR #541. The PR #541 shooting changes are unrelated.

Verification

Using Julia 1.11 and the official ModelingToolkitBase InterfaceII group with local BoundaryValueDiffEqCore, BoundaryValueDiffEqMIRK, and BoundaryValueDiffEqAscher:

• clean ModelingToolkit master at a35a814: BVP tests report 14 passed, 4 errored, 6 broken; the group fails
• this branch: BVP tests report 22 passed, 6 broken; the complete InterfaceII group passes in 3707.7 seconds

A bisect from 3c2e4a7 to a35a814 identified f71cc7d as the first bad commit.

The changed file passes Runic and git diff --check. Whole-repository Runic currently fails on unchanged master in src/systems/clock_inference.jl; that independent FormatCheck regression is intentionally not mixed into this focused PR.

[ChrisRackauckas-Claude]

Investigation scratchpad:

1. Read the failing BoundaryValueDiffEq master and PR #541 InterfaceII logs and matched all four errors to lib/ModelingToolkitBase/test/bvproblem.jl.
2. Reproduced the official downstream workflow locally against clean ModelingToolkit SHA a35a814 and BoundaryValueDiffEq SHA 219b502: 14 pass, 4 error, 6 broken in the BVP testset.
3. Ran git bisect between known-good 3c2e4a7 and failing a35a814; first bad commit was f71cc7d.
4. Restored the existing value and return-code assertions for the four successful solves.
5. Ran the complete ModelingToolkitBase InterfaceII group on this branch with the same local BoundaryValueDiffEq sublibraries: 22 pass, 6 broken in the BVP testset and 3707.7 seconds overall, exit 0.
6. Ran Runic on the changed file and git diff --check successfully. Whole-tree Runic also reproduces the independent clock_inference.jl failure on clean current master.

[ChrisRackauckas-Claude]

CI classification: the non-Runic reds are all present on the exact base commit 529c98930e before this PR. Master run 30363096231 has the same Catalyst, ModelingToolkitStandardLibrary, and SBMLToolkit failures, and master sublibrary run 30363097045 has the same ModelingToolkitBase QA failure (as well as the pre-existing InterfaceII failures this PR fixes). The failures are unrelated to this PR’s single @test_broken cleanup.

The Runic failure is also reproduced on clean master and is confined to src/systems/clock_inference.jl; a separate one-line explicit-return fix is under whole-repository Runic plus full InterfaceI validation. I am keeping that independent change out of this focused PR.

[ChrisRackauckas-Claude]

More specifically, the Catalyst job’s seven errors are all “Unexpected Pass” from stale @test_broken markers in conservation-law tests; focused Catalyst draft #1512 already activates exactly those seven assertions and has direct/full-group verification. That downstream fix is also independent of this ModelingToolkitBase test cleanup.

[ChrisRackauckas-Claude]

Independent clean-main audit after rebasing onto current master:

• Julia 1.12.6, current compatible BoundaryValueDiffEqMIRK 1.17.3 and BoundaryValueDiffEqAscher 1.15.2.
• Clean current main standalone Lotka–Volterra path: 0 pass, 2 errors. Both @test_broken solve(...) expressions returned valid numerical solution objects and Test reported “Expression evaluated to non-Boolean.”
• Executed adjacent bisect endpoints with the same dependency state: 05f590ee3 passed the original value assertions 4/4 in 5m08.2s; f71cc7d5 errored 2/2 in 5m07.7s with the same numerical solutions. git bisect therefore identifies f71cc7d5 as the first bad commit. This is a test regression, not a later solver/dependency regression.
• Rebased branch standalone bvproblem.jl: 22 passed, 6 pre-existing broken, exit 0. Repeated after the final rebase onto cdb7ad542; same 22/6 result, exit 0.
• Official GROUP=InterfaceII julia +1.12 --project=. -e "using Pkg; Pkg.test()": final BVP safetestset 22 passed / 6 broken; entire ModelingToolkitBase group passed, exit 0, 3940.14s.
• Runic 1.7.0 passes the changed file and git diff --check passes. Whole-tree Runic still reproduces the separate clean-master src/systems/clock_inference.jl missing-explicit-return failure; that remains outside this focused PR.

The lease-protected rebase push updated the branch from 7cbbc001f to tested commit cbd487c74 without overwriting any remote-only commit.