QA: mark the KLU JET assertion broken on the LTS, where Base logging is not inference-clean

[AJ0070]

Fixes #1190.

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

• One line of test annotation, no source change. The KLU assertion at test/qa/jet.jl:136 becomes broken = VERSION < v"1.11".
• Cause is in Base, not here. @SciMLMessage in the failure branches expands to Logging.@logmsg, and on 1.10 the Base.CoreLogging path it enters reaches Base.typejoin, which is itself a runtime dispatch on that version. Any solve that can emit a log therefore fails @test_opt on 1.10 no matter what LinearSolve does, and 1.11 carries the Base fix.
• Reproduced independently before changing anything: report_opt on this solve gives 2 reports on 1.10.11 and 0 on 1.12.6.
• The 1.10 reports name only typejoin, CoreLogging, _emit_log and SciMLLogging. None of them name getproperty or _extract!, which is what separates this from the KLU internals that JET: KLUFactorization solve is no longer dispatch-clean; getproperty self-recursion makes p, i, x, z = klu._L destructure a SparseMatrixCSC #1148 and KLU: extract factor parts through concrete helpers, not getproperty recursion #1163 fixed. This is not a regression of that work.
• broken rather than a skip, so it registers an unexpected pass and gets removed if the assertion ever starts holding on the LTS.
• Verified both directions rather than assuming the marker behaves: the assertion registers Pass on 1.12.6 and Broken on 1.10.11. That check matters because a marker that swallowed a real pass on 1.12 would quietly undo what KLU: extract factor parts through concrete helpers, not getproperty recursion #1163 established.
• Full test/qa/jet.jl on 1.12.6 is green with zero failures, and the sparse testset reads 1 pass and 2 broken as expected. Runic is clean.
• Note this is not currently a red CI check: the reusable workflow pins QA to QA_VERSIONS = ["1"], which overrides the ["lts", "1"] in test/test_groups.toml. It bites GROUP=QA runs on the LTS locally, and would become a CI failure if that list ever widened.

AI Disclosure: Used Opus 5