Fix Downgrade CI: raise MethodOfLines (0.11.12) and ModelingToolkit (11.27) floors

[ChrisRackauckas-Claude]

Ignore until reviewed by @ChrisRackauckas.

Problem

The Downgrade / Downgrade Tests lane on master is RED with 413 errored, 14 passed. julia-buildpkg succeeds (the floors resolve), but julia-runtest fails: nearly every mol_test.jl example errors inside MethodOfLines.discretize.

Root cause

At the declared floors the real julia-downgrade-compat@v2 resolution (StefanKarpinski/Resolver.jl, --min=@deps, julia=lts) lands on MethodOfLines 0.11.0-0.11.8, which pulls PDEBase 0.1.17. PDEBase 0.1.17 passes parameters to ODESystem as a Vector{Pair{Num,_}} (param => default):

ps = pdesys.ps === nothing ... ? Num[] : pdesys.ps
sys = ODESystem(alleqs, t, alldepvarsdisc, ps, ...)   # ps is Vector{Pair{Num,Float64}}

The MTK 9.x versions the resolver can co-select with MethodOfLines 0.11.8 (which needs Symbolics 6, i.e. MTK ≥ ~9.49) no longer accept pairs there:

• MTK 9.49-9.59 map getguess over ps, hitting getmetadata(::Pair, VariableGuess, nothing) — no method (this is the exact CI error).
• MTK 9.60+ reject the pairs in check_parameters: ArgumentError: v => 1 is not a parameter.

There is no resolvable MTK 9.x for the MethodOfLines 0.11.0-0.11.8 + PDEBase 0.1.17 + Symbolics 6 stack: MTK 9.x older than ~9.49 needs Symbolics 5 (incompatible with MethodOfLines 0.11.8), and ≥ 9.49 has the regression.

MethodOfLines 0.11.10+ requires MTK 11 and pulls PDEBase 0.1.20+, which uses the parameter API modern MTK expects — the same stack the green Core lts lane resolves to (MOL 0.11.12 / MTK 11.28.2 / PDEBase 0.1.25).

Fix

• MethodOfLines = "0.11.12, 0.12" — floor matches the green Core lane; pulls a PDEBase whose ps handling works with MTK 11.
• ModelingToolkit = "9.7.1, 10, 11.27" — MTK 11.0-11.26 declare julia = "1.9.0-1" but use Base.ispublic (Julia 1.11+), so they fail to precompile on the lts (Julia 1.10) downgrade target (UndefVarError: ispublic not defined). MTK 11.27 is the first 11.x whose compat correctly declares julia = "1.10.0-1".
• Removed the now-incorrect "expected RED until the advdiff3 test is fixed" comment in Downgrade.yml (the red was never about advdiff3).

Verification (local, Julia 1.10 lts)

Real julia-downgrade-compat resolution with the bumped floors lands on: MethodOfLines 0.11.12, ModelingToolkit 11.27.0, PDEBase 0.1.25, SciMLBase 3.18.0, Symbolics 7.28.1, OrdinaryDiffEqSDIRK 2.0.0, NonlinearSolve 4.17.1, OrdinaryDiffEq 7.0.0.

Running test/mol_test.jl against that pinned stack: all 427 examples run, 0 errored, 0 failed, 1 broken (adv3, intentional @test_broken). convsin — the first error in CI — now passes in 4.0s. Before: 14 passed, 413 errored.

🤖 Generated with Claude Code