Fix QA group dispatch: pass QA via the qa kwarg, not groups

[ChrisRackauckas-Claude]

Problem

The QA CI job fails on master (introduced by the canonicalization conversion #154) before any test runs:

ERROR: LoadError: ArgumentError: run_tests: GROUP="QA" was requested but no `qa` body was provided
   @ SciMLTesting .../SciMLTesting.jl:959
   @ test/runtests.jl:3

The conversion placed the QA group inside the groups Dict. But SciMLTesting.run_tests routes GROUP="QA" through a dedicated elseif group == "QA" branch that consumes the qa keyword argument and short-circuits before the groups table is ever consulted. With qa left unset, that branch throws "no qa body was provided" — so a QA entry in groups is unreachable for GROUP="QA".

Fix

Move the QA spec from the groups Dict to the dedicated qa keyword argument (the documented SciMLTesting form). all = ["Core"] is retained, so QA still runs only when GROUP="QA" is explicitly selected and remains excluded from the default GROUP="All" / local ]test run, exactly as before.

Verification (local, Julia 1.10 + SciMLTesting v1.2.0)

Reproduced the exact CI error with the old groups["QA"] form, then confirmed the fix with a stub tree mirroring test/ (core.jl + qa/qa.jl + qa/Project.toml):

• GROUP=QA → exit 0, QA body runs (was: ArgumentError: no qa body provided)
• GROUP=Core → exit 0, Core only
• GROUP=All / no GROUP → exit 0, Core only (QA excluded)

In a checkout of this repo, GROUP=QA julia test/runtests.jl now reaches the QA branch, activates test/qa, develops local FEniCS, and resolves Aqua/JET/PyCall — it no longer throws the dispatch error (the local run then needs the cmhyett/julia-fenics Python stack, which is only present in CI).

Note: the other three master jobs (Core julia 1, Core julia pre, QA julia lts) failed separately with permission denied ... docker.sock — transient self-hosted-runner infrastructure errors unrelated to this change; not addressed here.

Please ignore until reviewed by @ChrisRackauckas.

---

Follow-up: fix the QA package-quality failures the QA group now surfaces (commits 85a709b, f70877a)

With the dispatch fixed, the QA group runs FEniCS's own Aqua.test_all/JET.report_package for the first time, exposing six pre-existing package-quality issues (all present on master, none introduced here). Each is fixed at its real cause; no Aqua check is disabled or loosened:

QA check	Finding	Fix
Piracy	write(path, solution, time::Number) (jsolve.jl) extended Base.write with no FEniCS-owned argument type.	Keep extending Base.write (so write stays a single function usable unqualified), but type solution::fenicsobject so the method is FEniCS-owned, not piracy. The unused, unavoidably-piratical write(::Any, ::PyObject, ::Number) overload is dropped (use store for raw vectors).
Method ambiguity	The write method's untyped path clashed with Base.write(::IO, ...) and write(::AbstractString, ...).	Type path::PyObject (what XDMFFile/TimeSeries return), disjoint from IO/AbstractString.
Undefined exports	export fenicsclass named a binding that doesn't exist — the macro is @fenicsclass.	export @fenicsclass.
JET (1 report)	Argument(V, number) → Argument(::Any, ::Any, ::Nothing) no matching method, because part::StringOrSymbol cannot hold its own nothing default.	Widen to part::Union{StringOrSymbol, Nothing} = nothing (PyCall maps nothing→None, matching fenics.Argument(part=None)).
Stale dependencies	ProgressMeter declared but unused (only a no-op @require ... using ProgressMeter; no function from it is ever called).	Remove from [deps], [compat], and the dead __init__ block.
deps_compat	LinearAlgebra (runtime stdlib dep) had no [compat] entry.	Add LinearAlgebra = "1".

The first attempt (85a709b) instead removed write from the Base import, which made FEniCS.write clash with the exported Base.write and broke the Core examples (UndefVarError: write in tutorial8/9). f70877a corrects that by extending Base.write and disambiguating via argument types as above.

Verified locally in the cmhyett/julia-fenics container (the fenics Python stack) with Julia 1.10: Aqua.test_all(FEniCS) → 11 passed / 0 failed; JET.test_package(FEniCS) → 1 passed / 0 failed; and a module … using FEniCS; write(…) sanity check confirms the Base.write name clash is gone. (The earlier CI on 85a709b had already shown both QA legs green — i.e. piracy/exports/stale/deps_compat/JET all fixed — and Core red from the clash; f70877a keeps QA green and unbreaks Core.)

Please ignore until reviewed by @ChrisRackauckas.