Restore the scalar DAE Jacobian test helper

[ChrisRackauckas-Claude]

Please ignore this PR until reviewed by @ChrisRackauckas.

What changed and why

Return the scalar Jacobian value from the out-of-place DAE convergence-test helper. A Runic cleanup converted its previous value-returning assignment into a bare return, so downgrade CI passed nothing to lu before reaching the convergence assertion.

An exact-action reproduction and source-history check identified 09173df as the introducing commit.

Failing before / passing after

The existing downgrade test discriminates without adding or weakening an assertion. On unfixed master:

DAE Convergence Tests | Pass 45 Error 1 Total 46
MethodError: no method matching lu(::Nothing; check::Bool)

With the one-line return restored, the identical command passes the full BDF Core group:

DAE Convergence Tests | 50/50
DAE AD Tests | 40/40
DAE Event Tests | 4/4
DAE derivative_discontinuity! Tests | 11/11
DAE Initialization Tests | 38 passed, 2 existing broken
DAE Nonlinear Solve Path Tests | 16/16
BDF Inference Tests | 1/1
BDF Convergence Tests | 84/84
BDF Regression Tests | 33/33
Nordsieck BDF Tests | 101/101
Testing OrdinaryDiffEqBDF tests passed

Local verification

• Exact downgrade action on Julia 1.11: TMPDIR="$PWD/.ci-tmp" JULIA_DEPOT_PATH="$PWD/.ci-depot" JULIA_NUM_PRECOMPILE_TASKS=2 GROUP=Core julia +1.11 --startup-file=no --color=no --project=lib/OrdinaryDiffEqBDF -e 'using Pkg; Pkg.test(; allow_reresolve = false)': full group passed after the 45-pass/1-error failure above.
• GROUP=QA julia +1.12 --startup-file=no --project=lib/OrdinaryDiffEqBDF -e 'using Pkg; Pkg.test()': Allocation 2 pass/18 pre-existing broken, JET 18 pass/23 pre-existing broken, Aqua 21/21; package tests passed.
• Runic, typos, and git diff --check: passed.

Docs, GPU, downstream, and prerelease-Julia groups were not run because this only repairs a test helper used by the BDF Core/downgrade group.

[ChrisRackauckas-Claude]

CI triage: all completed failures are the two independently reproduced master Downstream issues. ProbNumDiffEq’s Julia 1.10 resolver failure is addressed by #4320. PositiveIntegrators’ LinearSolve 5 compatibility failure is addressed upstream by NumericalMathematics/PositiveIntegrators.jl#216. No completed red check currently points to this PR’s changed behavior.