QA: restore JET execution + remove broken duplicate QA.yml

[ChrisRackauckas-Claude]

Summary

CI centralization (#141) wired a new [QA] test group but left two regressions on master:

1. Broken duplicate workflow. The standalone .github/workflows/QA.yml was not removed by Canonical CI: grouped-tests.yml + root test/test_groups.toml #141. It still runs julia --project=test/qa test/qa/runtests.jl, but Canonical CI: grouped-tests.yml + root test/test_groups.toml #141 deleted test/qa/runtests.jl. So the QA - Julia lts / QA - Julia 1 jobs are red on master, and they duplicate the new [QA] group (which runs via the grouped-tests.yml@v1 caller in Tests.yml, dispatched from test/test_groups.toml).

2. A passing enforced check was silently dropped. The old QA.yml enforced a real JET analysis — JET.report_package(MultiScaleArrays; target_modules=(MultiScaleArrays,)) with @test length(reports)==0 — and it passed on master. The new test/qa/qa.jl ran no JET at all, only @test_broken false.

Fix

• Delete .github/workflows/QA.yml — the [QA] group supersedes it; this removes the broken, duplicate jobs.
• Restore the real JET check in test/qa/qa.jl, enforced with @test (not skipped). The Aqua @test_broken markers (ambiguities + deps_compat, issue QA: Aqua/JET findings marked @test_broken pending fix #142) are left unchanged — they are a separate, still-open concern and out of scope here.

Local verification (julia +1.11, GROUP=QA)

The test/qa env resolves the same stack the old standalone env did (DiffEqBase 7.5.5, JET 0.9.20). The restored check reports 0 findings under target_modules=(MultiScaleArrays,):

Test Summary: | Pass  Total   Time
JET           |    1      1   34.5s

Full QA group via Pkg.test() with GROUP=QA (exit 0):

Test Summary:     | Pass  Broken  Total     Time
Quality Assurance |    7       2      9   1m25.3s
     Testing MultiScaleArrays tests passed

The 2 Broken are the pre-existing Aqua markers from #142; 0 Fail / 0 Error.

Note on the #142 JET finding

The DiffEqBase.alg_needs_extra_process finding listed in #142 comes from the stricter JET.test_package(target_defined_modules=true), not from the check that was actually dropped. With target_modules=(MultiScaleArrays,), JET attributes that undefined-binding to the DiffEqBase virtual-toplevel frame and the module filter drops it (confirmed locally: report_package without target_modules does surface it, 3 reports; with the filter, 0). This PR restores the historical enforced check verbatim — it is not a test-logic change. The src/diffeq.jl:427 source bug (using DiffEqBase. instead of StochasticDiffEq.) is real and still tracked in #142 for a source-level fix.

Ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code