Uniformize monorepo structure to OrdinaryDiffEq.jl canonical style

[ChrisRackauckas-Claude]

Uniformizes Optimization.jl to the canonical OrdinaryDiffEq.jl monorepo structure.

Canonicalization done

OptimizationBase test/Project.toml removed (the only forbidden main test/Project.toml)

• Folded its [deps] into lib/OptimizationBase/Project.toml [extras] + [targets].test.
• Moved its [sources] (OptimizationLBFGSB, OptimizationManopt) into the package [sources], corrected to sibling paths ../OptimizationLBFGSB / ../OptimizationManopt (they were ../../ because they previously lived under test/).
• Added matching [compat] for the new test extras (Aqua, ComponentArrays, IterTools, Lux, Manifolds, ModelingToolkit, Optim, Optimisers, OptimizationLBFGSB, OptimizationManopt, SafeTestsets, Symbolics, Tracker, BenchmarkTools, ChainRulesCore, Random, Pkg, SparseArrays, Test). No [compat] floor was lowered.
• Names already present in [deps] (ADTypes, DifferentiationInterface, LinearAlgebra, SciMLBase, SparseArrays, SparseConnectivityTracer) are kept only in [targets].test, not duplicated into [extras].

Group threading (OPTIMIZATION_TEST_GROUP)

• Every lib/<X>/test/runtests.jl is now a SafeTestsets group dispatcher: it reads get(ENV, "OPTIMIZATION_TEST_GROUP", "ALL") and @safetestset-includes the package's existing functional suite (moved verbatim to core_tests.jl) under the Core/ALL group, mirroring the OrdinaryDiffEq sublib runtests.jl pattern.
• SafeTestsets added to each sublib's [extras] / [compat] / [targets].test.

Root test/runtests.jl

• Rewritten as the canonical _detect_sublibrary_group dispatcher: handles bare sublib names and <sublib>_<group> suffixes, transitively Pkg.develops a sublib's [sources] on Julia < 1.11, sets OPTIMIZATION_TEST_GROUP, and Pkg.tests the sublib; otherwise runs the root's own Core/GPU @safetestset groups. Fixes the previously-undefined activate_downstream_env in the GPU branch.

Workflows

• SublibraryCI.yml: passes group-env-name: OPTIMIZATION_TEST_GROUP and check-bounds: auto to sublibrary-project-tests.yml@v1.
• DowngradeSublibraries.yml: passes group-env-name/group-env-value: Core, extends skip with the locally-developed in-repo sibling packages (OptimizationBase + the seven other [sources] siblings) plus stdlibs and SymbolicAnalysis. No allow-reresolve (removed in v1).
• Root CI.yml (tests.yml@v1, GROUP-dispatched to Core) and Downgrade.yml (downgrade.yml@v1, no allow-reresolve) were already canonical and contain no hardcoded sublibrary GROUP rows (sublibs are covered by the SublibraryCI project model).

test_groups.toml

• Added a Core-only test/test_groups.toml to every sublib so the matrix discovery declares the real group (Core on [lts, 1.11, 1, pre], matching the julia 1.10 floor) and does not spawn a phantom default QA-on-[1] job — no sublib has a QA suite (the Aqua extras in Ipopt/MadNLP are unused by their runtests.jl).

Root [sources] justification (invariant B)

The root Optimization package re-exports only OptimizationBase (its sole in-repo [deps] dependency); the solver sublibraries are not root dependencies. Root [sources] therefore lists OptimizationBase plus the four solver sublibraries used by the root test target (OptimizationLBFGSB, OptimizationMOI, OptimizationOptimJL, OptimizationOptimisers). Listing all 27 would inject false dependencies and break the true path graph, so it is intentionally left at 5.

Verified locally (isolated depot, Julia 1.11)

• TOML-parsed every changed Project.toml + test_groups.toml: all parse; every [targets].test entry resolves via [extras]∪[deps]; no name appears in both [deps] and [extras]; zero test/Project.toml remain (only the canonical test/downstream/Project.toml sub-environment); zero @testitem/ReTestItems anywhere.
• YAML-parsed all changed workflows: valid.
• Instantiated two representative sublibs: OptimizationBBO and OptimizationPolyalgorithms (the latter resolving all three in-repo siblings via [sources]).
• Ran OptimizationBBO Core group through the new @safetestset dispatcher: Core | 25 25 pass, Testing OptimizationBBO tests passed.
• Unit-tested the root _detect_sublibrary_group routing (OptimizationBBO→sublib/Core, OptimizationBBO_QA→sublib/QA, Core→root-Core, GPU→root-GPU): all correct.
• Ran Runic on every authored .jl (28 dispatchers + root runtests): already clean, 0 reformats.

Left to CI

• The full per-sublibrary SublibraryCI matrix (26 of 27 sublibraries were not run locally; many need heavy/external solvers — Ipopt, MadNLP, NLopt, Python via SciPy/PyCMA, etc.).
• The root Core group end-to-end via Pkg.test (heavy root test env: Enzyme, Lux, SciMLSensitivity, ...) and the Downgrade/DowngradeSublibraries resolves.

Ignore until reviewed by @ChrisRackauckas.