Fix QA test group: pass qa body via dedicated run_tests kwarg

[ChrisRackauckas-Claude]

Problem

Both tests / QA jobs (julia 1 and julia lts) fail on master at test startup with:

ERROR: LoadError: ArgumentError: run_tests: GROUP="QA" was requested but no `qa` body was provided

Root cause

SciMLTesting 1.2's run_tests routes the reserved "QA" group through a dedicated qa= keyword argument, not through the groups Dict. In run_tests, the elseif group == "QA" branch is evaluated before groups is consulted, and it throws when qa === nothing:

elseif group == "QA"
    qa === nothing &&
        throw(ArgumentError("run_tests: GROUP=\"QA\" was requested but no `qa` body was provided"))
    _run_group_spec(_group_spec(qa), parent; label = "QA")

test/runtests.jl declared the QA spec inside groups = Dict("QA" => ...) and left qa unset, so the dispatch errored before it ever reached the groups table. This was introduced in the v1.2 folder-model conversion (#619 normalized the group into canonical QA, but the wiring was still via groups).

The reusable CI workflow (grouped-tests.yml@v1) reads test/test_groups.toml, which correctly declares the QA group on lts/1, and dispatches GROUP=QA to Pkg.test. The workflow and test_groups.toml are correct; only runtests.jl was miswired.

Fix

Move the QA spec from groups = Dict("QA" => (; env, body)) to the dedicated qa = (; env, body) keyword, matching the SciMLTesting v1.2 explicit-args API. No other change.

Verification (local)

Against SciMLTesting 1.2.0 on Julia 1.10:

• Old shape reproduces the CI error: passing QA in groups with GROUP=QA throws the exact ArgumentError: GROUP="QA" was requested but no qa body was provided.
• New shape works: with qa = (; ...), GROUP=QA reaches the QA body (new QA body reached), and GROUP=Core still reaches the core body — no error.
• Runic.format_string confirms test/runtests.jl is already correctly formatted (FormatCheck clean).

Please ignore until reviewed by @ChrisRackauckas.

[ChrisRackauckas-Claude]

Additional fix on this branch: docs build (@mtkmodel removed in MTK v11)

Besides the QA-group wiring, the Documentation job was red on a separate genuine bug. The @example block in docs/src/libs/datadrivensparse/example_04.md (autoregulation) failed with:

UndefVarError: `@mtkmodel` not defined in `Main.__atexample__named__example_04`

Root cause

ModelingToolkit v10/v11 removed the @mtkmodel macro entirely — it is no longer defined or exported (isdefined(ModelingToolkit, Symbol("@mtkmodel")) == false on MTK v11.26.8; only @mtkbuild/@mtkcompile/@mtkcomplete remain). The example had been partially ported (it already used @mtkcompile) but still defined the model with @mtkmodel, so the docs build broke. (The triage hint that this was a "missing using ModelingToolkit" / same-block macro issue was incorrect — using ModelingToolkit was already present; the macro simply does not exist anymore.)

Fix

Replace the @mtkmodel/@mtkcompile model with the MTK v11 direct System construction:

@parameters α = 1.0 β = 1.3 γ = 2.0 δ = 0.5
@variables (x(t))[1:2] = [20.0, 12.0]
eqs = [
    D(x[1]) ~ α / (1 + x[2]) - β * x[1],
    D(x[2]) ~ γ / (1 + x[1]) - δ * x[2],
]
@named autoregulation = System(eqs, t)
sys = mtkcompile(autoregulation)
de_problem = ODEProblem(sys, [], tspan)

This yields the same sys.x array variables the rest of the example consumes. Also dropped the unused NoSpecialize import and the ODEProblem{true, NoSpecialize} type params (plain ODEProblem(sys, [], tspan) is the v11 form).

Verification (local, Julia 1.12 / ModelingToolkit v11.26.8)

A focused Documenter build of the example_04 page (with :example_block NOT in warnonly, so any block failure errors the build) ran every @example block with zero failed to run warnings and makedocs completed successfully (VERIFY_MAKEDOCS_OK, exit 0).

Please ignore until reviewed by @ChrisRackauckas.