QA: run_qa v1.6 form + ExplicitImports

[ChrisRackauckas-Claude]

This PR should be ignored until reviewed by @ChrisRackauckas.

Brings this repo's QA group onto the SciMLTesting v1.6 run_qa form with ExplicitImports enabled, converting test/qa/qa.jl from hand-rolled per-check Aqua calls.

ExplicitImports findings (6 checks)

Check	Result	Action
no_implicit_imports	broken	37 implicit names from heavy whole-module using (CUDA/Flux/Zygote/DiffEqBase-reexport/etc.); making them explicit is a deliberate refactor. ei_broken = (:no_implicit_imports,), tracked in #155 (auto-flags once resolved).
all_qualified_accesses_via_owners	fixed	AbstractODEAlgorithm / AbstractODEProblem were accessed via DiffEqBase.<name> but are owned by SciMLBase (DiffEqBase only reexports them). Switched the three subtype annotations in src/HighDimPDE.jl to SciMLBase.<name>.
all_qualified_accesses_are_public	ignored	6 non-public names from other packages (SciMLBase: AbstractODEAlgorithm, AbstractODEProblem, AbstractSciMLProblem; Flux/Flux.Optimise: AbstractOptimiser, Optimise, params). Added to the check's ignore list with source pkg documented; these go public as the base libs add public/exports.
no_stale_explicit_imports	pass
all_explicit_imports_via_owners	pass
all_explicit_imports_are_public	pass

Also: pre-existing Aqua stale_deps fix

Test was in the package [deps] but is never imported by src/ — a genuine stale dep (this failure reproduces on unmodified main with the old qa.jl too). Test is a test-only stdlib already present in [extras]/[targets].test; removed it from [deps].

Deps

• SciMLTesting [compat] floor bumped to "1.6" in root and test/qa/Project.toml.
• ExplicitImports stays transitive via SciMLTesting (not added directly). Aqua kept (ambiguities sub-check needs it as a direct dep). No JET (not used).

Local verification (Julia 1.11, released SciMLTesting 1.6.0)

Test Summary:     | Pass  Broken  Total     Time
Quality Assurance |   16       1     17  1m41.9s

0 Fail / 0 Error. The 1 Broken is the intended no_implicit_imports placeholder.

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Fixed the red QA (julia 1) lane.

Root cause (verified locally, not introduced by this conversion): on Julia 1.12 Aqua test_undefined_exports flags du_cache/u_cache/user_cache. These are owned by SciMLBase, re-exported by DiffEqBase, then re-exported again here via @reexport using DiffEqBase. On Julia <=1.11 the binding resolves (isdefined == true, Aqua passes); Julia 1.12's binding-partition change makes the double-reexported names report isdefined == false, so they look like undefined exports. They are absent from src/ — purely a 1.12 artifact of the SciMLBase->DiffEqBase reexport chain.

Pre-existing: reproduces on unmodified main (scheduled run 28211297790, job 83573031987 — the old hand-rolled qa.jl already ran Aqua.test_undefined_exports). This conversion didn't introduce it; it actually fixed the stale_deps failure that main also had on that lane.

Fix: aqua_broken = (:undefined_exports,) with a comment + tracking issue #157. run_qa disables the sub-check in Aqua.test_all and emits a tracked @test_broken; remove the entry once the upstream reexport chain is fixed on 1.12.

Local verification (released SciMLTesting 1.7.0):

• Julia 1.12.6 (the gated lane): Quality Assurance | 15 Pass, 2 Broken, 0 Fail, 0 Error (the 2 broken = undefined_exports + no_implicit_imports placeholder).
• Julia 1.10: confirmed du_cache etc. resolve (Undefined exports: Symbol[]), i.e. the failure is genuinely 1.12-only.

Note: CI only schedules QA (julia 1); there is no QA (lts) lane (grouped-tests.yml@v1).