Use released downstream workflow v1

[ChrisRackauckas]

Summary

• Follow up on Use the shared downstream workflow on Julia 1 #4323 by replacing the temporary SciML/.github merge-commit pin with the released moving v1 tag.
• SciML/.github v1.28.0 and v1 both resolve to d5748d7f7e95749cb0a5acaed6e69162623be862, so this changes the reference without changing the reusable workflow content.
• Keep upstream-subdirs: ".,lib/*" unchanged.
• Apply the two indentation changes required by Runic 1.9.0, which current master also reports.

Verification

• actionlint 1.7.12 on .github/workflows/Downstream.yml: passed.
• typos on both changed files: passed.
• git diff --check: passed.
• Confirmed all twelve SciML reusable workflow calls in this repository now use @v1.
• Ran the real GROUP=QA Pkg.test() on Julia 1.12.7 with fresh unconstrained resolution to GenericSchur 0.5.8: 92 of 92 passed.
• Ran Runic 1.9.0 over the full repository: passed.

Initial CI context

The first CI run resolved GenericSchur 0.5.7 and hit GenericSchur #24 while precompiling ExponentialUtilities. The identical failure was already present on current master CI. GenericSchur #25 has since merged, and v0.5.8 is released and registered.

No dependency workaround is included in this branch. The follow-up push reruns CI against the patched release.

[ChrisRackauckas]

Implementation record for commit 488135360043589956b2bfdfad0399e7c309ed70:

1. Released SciML/.github v1.28.0 from merge commit d5748d7f7e95749cb0a5acaed6e69162623be862.
2. Verified the release workflow completed and moved v1 to that same commit.
3. Replaced the temporary exact-SHA reference in Downstream.yml with @v1.
4. Verified the workflow with actionlint 1.7.12, typos, and git diff --check.
5. Ran the unchanged QA tests in a disposable Julia 1.12.7 environment with GenericSchur held at the last non-regressed version, 0.5.6. All 92 QA tests passed.

The unconstrained dependency failure and upstream evidence are recorded in the PR description. No dependency workaround is included in this branch.

[ChrisRackauckas]

Follow-up for commit 2e8038c2e87ffd8cf5c35d956ce0b9941440935a:

• The initial run had 38 failures, and every failed check name matched current master. Thirty-four were the GenericSchur 0.5.7 regression; the remaining four were existing PositiveIntegrators, Runic, BDF downgrade, and NonlinearSolve downgrade failures.
• GenericSchur Fix resize! methods #25 has now merged, and v0.5.8 is released and present in General.
• Applied the two exact indentation changes requested by Runic 1.9.0 in lib/OrdinaryDiffEqSDIRK/src/generic_imex_perform_step.jl.
• Reran full-repository Runic: passed.
• Reran the real unpinned GROUP=QA Pkg.test() on Julia 1.12.7, which resolved GenericSchur 0.5.8: 92 of 92 passed.
• Reran actionlint, typos, and git diff --check: passed.

This push triggers a fresh CI run against the patched dependency release.

[ChrisRackauckas]

Final CI audit for head 2e8038c2e87ffd8cf5c35d956ce0b9941440935a:

• GitHub's terminal rollup is 126 successful, 14 failed, 3 skipped, 0 active, and 0 cancelled.
• GenericSchur 0.5.8 removed the initial geigen! precompile cascade. None of the refreshed failures has that signature.
• The released @v1 workflow expanded all 59 upstream projects and completed the full downstream matrix. SciMLSensitivity Core3, the longest job at about 3 hours 18 minutes, passed.
• No refreshed failure is attributable to the @v1 reference.

The 14 failures are classified as unrelated current baseline or downstream release defects:

• 3 recurring InterfaceV JLArray pointer failures.
• 3 existing BDF test-dependency or metadata failures.
• 2 SciMLSensitivity Adam failures after OptimizationOptimisers 0.3.22 stopped re-exporting it, confirmed by Optimization.jl #1308.
• 1 PositiveIntegrators LinearSolve compatibility conflict, identical to Use the shared downstream workflow on Julia 1 #4323 and master.
• 1 existing OrdinaryDiffEqNonlinearSolve downgrade failure.
• 1 recurring Newton-Krylov roundoff failure.
• 1 recurring GPU mismatch failure.
• 1 ModelingToolkit NelderMead failure after OptimizationOptimJL 0.4.19 changed re-exports, tracked by Optimization.jl #1315.
• 1 SciMLSensitivity Core1 Enzyme runtime-activity failure, identical to current SciMLSensitivity master.

The PR remains open and mergeable. I did not add unrelated dependency or test-metadata workarounds to this branch.