Target downstream tests in the OrdinaryDiffEq monorepo

[ChrisRackauckas-Claude]

Please ignore this PR until reviewed by @ChrisRackauckas.

Summary

• replace the archived standalone DiffEqBase and DelayDiffEq targets with their current OrdinaryDiffEq monorepo subdirectories
• keep OrdinaryDiffEq's root Interface job, while removing the stale duplicate root Core row
• forward the matrix subdirectory to the reusable downstream workflow

This branch includes the LabelledArrays group correction already proposed in #632. Merge #632 first; until then GitHub will show that prerequisite line in this stacked diff.

Depends on:

• Support monorepo package projects in downstream CI .github#115 for reusable-workflow subdir support
• Honor generic downstream test groups in migrated subpackages OrdinaryDiffEq.jl#3909 so migrated sublibrary runners honor the reusable workflow's generic GROUP

Local verification

• actionlint .github/workflows/Downstream.yml: passed
• julia +1.12 -m Runic --check .: passed
• git diff --check: passed
• exact DelayDiffEq Interface target: package tests passed; representative summaries include AD 15 pass/6 broken, Backwards 16/16, Composite Solution 12/12, History Function 81/81, and Multi-Algorithm 28/28
• exact DiffEqBase Downstream2 target: ran substantive tests (Prob Kwargs 3/3, Unwrapping 3/3, DE stats 4/4) before reaching the independently reproduced obsolete VectorContinuousCallback call in community_callback_tests.jl:230; that clean-base failure is being fixed separately
• generic-group DelayDiffEq Interface verification against the OrdinaryDiffEq fallback change: package tests passed

The target/group audit classifies every row in the proposed matrix as a real selected group; none falls back silently.

[ChrisRackauckas-Claude]

The clean-base DiffEqBase Downstream2 callback failure exposed here has now been bisected and fixed in SciML/OrdinaryDiffEq.jl#3914. The first bad SciMLBase migration commit was ba118add; the fix restores the old upcrossing-only event behavior with the current three-argument VCC constructor. Local Julia 1.12 verification passed the targeted solve and the full Downstream2 group (Community 16/16, Distributed 4/4, package exit 0).

[ChrisRackauckas-Claude]

CI timer check (2026-07-15 09:54 EDT): Documentation, Downgrade, Downgrade Sublibraries, Sublibrary CI, formatting, spelling, and Runic all passed. Every ordinary CI matrix job has also passed; only the self-hosted GPU job remains queued.

IntegrationTest has the expected startup failure because this caller uses the new upstream-subdirs input that is not yet present in the shared @v1 workflow. It can start after SciML/.github#115 is merged and the v1 tag is advanced; the exact local downstream rehearsals are recorded in the PR body. The branch remains current with upstream/master.

⎿ Setting 4-hour timer for CI completion as per CLAUDE.md instructions...
Current time: 2026-07-15 09:54 EDT
Will check CI results at: 2026-07-15 13:54 EDT

[ChrisRackauckas-Claude]

The previous IntegrationTest run (29409776792) was rejected with startup_failure before the shared SciML/.github@v1 workflow release. I pushed empty commit e4cf7f4 to retrigger it against the released reusable workflow; no source or workflow content changed in that commit.

⎿ Setting 3-hour timer for CI completion as per CLAUDE.md instructions...
Current time: 2026-07-16 13:06:58 EDT
Will check CI results at: 2026-07-16 16:06:58 EDT

[ChrisRackauckas-Claude]

Final integration-runtime check for run 29518442731: all 13 downstream matrix rows completed successfully. Every completed row entered its selected test group and emitted substantive test evidence; the last row, SciMLSensitivity Core3, emitted a Test Summary: and the package-level “SciMLSensitivity tests passed” marker before completing. This confirms the monorepo/subdirectory routing changes run tests rather than silently falling back or exiting early.