Move the SupernodalLU sweep proof to a QA-only file

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas.

Follow-up to #1119, which merged while this was in progress — rebased onto main. Addresses the structural half of the macOS failure: the pre job specifically.

The problem

test/runtests.jl includes qa/allocations.jl from two groups that do not cover the same Julia versions:

[QA]              versions = ["lts", "1"]
[AppleAccelerate] versions = ["lts", "1", "pre"]

So AllocCheck ran on pre through a group that was never meant to carry it.

The split

The file mixes two things with different portability:

• Direct-BLAS refactorization proofs — hold on every release. That contract is the reason the AppleAccelerate group exists. Stay put, still run under both groups on all versions.
• The SupernodalLU sweep proof — above PANEL_BLAS_CUTOFF the sweeps hand off to LinearAlgebra's ldiv!/mul!, whose wrappers leave a generic_trimatdiv! dynamic dispatch plus boxed MulAddMul/SubArray values that AllocCheck reports on 1.10 and 1.13. Moves to test/qa/supernodal_allocations.jl, included only by QA.

Measured on the same two matrices the testset uses:

Julia	arch	static findings	runtime 1-RHS	3-RHS
1.10.11	x86_64-Linux	18	0 B	0 B
1.12.6	x86_64-Linux	0	0 B	0 B

18 findings on 1.10 x86_64-Linux is the same count #1119 measured on 1.10 aarch64-Darwin, and 1.12 is clean on both — so this is the stdlib and the Julia version, not the platform and not this package's code. (That also closes the "I have no Linux box here" caveat in #1119.)

Worth flagging: the coverage is narrower than the toml suggests — only one QA job actually materialises (QA (julia 1)), so the static proof is exercised on 1.12 alone, and it was green on Linux only because QA/lts never runs.

One deliberate difference from #1119

The static proof keeps a version guard — that is a genuine limit of the prover, and the runtime numbers above show the sweeps are allocation-free anyway.

RUNTIME_MULTIRHS_ZERO is dropped. It gated a runtime allocation assertion, and per #1119's own table the 1.11 multi-RHS path allocates 6144 B for real — that is a regression, not a prover limitation, and skipping the assertion hides it. With the gate gone the test fails on 1.11 until the allocation is fixed, which I think is the correct behavior. 1.11 is not in the CI matrix (lts/1/pre = 1.10/1.12/1.13), so this costs nothing today. I could not reproduce the 6144 B myself — my 1.11 QA env failed to precompile AllocCheck — so I have not verified that number independently.

Testing

• qa/allocations.jl completes on Julia 1.10, the version that was failing.
• GROUP=QA on Julia 1.12 (what CI runs): rc=0.
• GROUP=Core on Julia 1.12: rc=0.
• Runic reports no formatting changes.

🤖 Generated with Claude Code

https://claude.ai/code/session_017rr42T1vFRRT8D7DMAmJzf