Target the DiffEqDevTools monorepo test project

[ChrisRackauckas-Claude]

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

Audit finding

The downstream workflow still cloned the archived standalone DiffEqDevTools.jl repository and selected All. DiffEqDevTools is now registered from OrdinaryDiffEq.jl:lib/DiffEqDevTools, whose focused group is Core and whose direct runner reads ODEDIFFEQ_TEST_GROUP. The workflow also caught resolver errors and converted them to success.

Changes

• clone OrdinaryDiffEq.jl and activate lib/DiffEqDevTools;
• select the populated Core group through ODEDIFFEQ_TEST_GROUP;
• keep the existing root OrdinaryDiffEq Regression_I row;
• activate and test the selected project explicitly;
• develop all DiffEqProblemLibrary subpackages into the downstream environment;
• let resolver and test failures propagate.

Local verification

• actionlint passed for .github/workflows/Downstream.yml.
• Runic.jl repository check passed.
• OrdinaryDiffEq/Regression_I: exit 0; OrdinaryDiffEq tests passed.
• Workflow-equivalent Julia 1.10.11 DiffEqDevTools/Core: exit 0. Final sets included Plot Recipes 36/36 and nonlinear-solve work-precision diagrams 6/6; Pkg reported DiffEqDevTools tests passed.
• git diff --check passed.

An earlier All run exposed unrelated pre-existing QA/documentation coverage failures; those are being handled separately in SciML/OrdinaryDiffEq.jl#3906. This workflow now requests only the intended populated Core group.

[ChrisRackauckas-Claude]

The remaining OrdinaryDiffEq.jl/Regression_I/1.10 failure (job 87315053265) is a self-hosted runner loss, not a test assertion. It ran on self-hosted-4vcpu-8gb-vv9dn-runner-hmgds; the downstream test step began at 09:41:32Z and was still marked in_progress when the job failed at 11:01:55Z, with subsequent steps pending and the standard runner-lost annotation. This evidence is being added to SciML/.github#116.