Fix Downgrade (Core) and Aqua/JET QA failures

[ChrisRackauckas-Claude]

Summary

Fixes the Downgrade (Core) check and the Aqua method-ambiguity / unbound-type-parameter / deps-compat and JET sub-checks of the QA jobs on main. All fixes were reproduced and verified locally on the failing Julia versions before committing.

1. Downgrade (Core) — SteadyStateDiffEq compat floor

The downgrade job failed precompiling SteadyStateDiffEq with UndefVarError: AbstractNonlinearTerminationMode not defined (its solve.jl:70 references SciMLBase.AbstractNonlinearTerminationMode). The old floor SteadyStateDiffEq = "1, 2" let the downgrade resolver pick a SteadyStateDiffEq paired with a SciMLBase too old to define that symbol. Raised to SteadyStateDiffEq = "2.5", which forces a SciMLBase floor (resolved 2.99.0) that defines it. Also added the missing SparseArrays = "1" compat (fixes the Aqua deps-compat failure) and raised the DiffEqBase floor to 6.140 for downgrade resolvability.

2. QA / Aqua — method ambiguity + unbound type parameters

pairedindices(::DefaultIndexHandler{N}, ::NTuple{N,T}, ::CartesianIndex{N}) had a T<:Number/T<:AbstractVector pair that was ambiguous at N == 0 (NTuple{0,T} === Tuple{} leaves T free) and an AbstractVector fallback with unbound M/T. Rewrote the two methods using Tuple{T, Vararg{T}} (so T is always bound) plus an explicit zero-dimensional method, and dropped the unused type params on the dimension-mismatch fallback.

3. QA / JET — NullParameters

NullParameters is not exported from DiffEqBase, so the bare default-arg references in build_rhs.jl / build_rhs_ss.jl were undefined bindings. Qualified as DiffEqBase.NullParameters(). This also clears the JET iterate(::Nothing) cascade that stemmed from the ambiguous/unbound pairedindices methods.

Verification (run locally)

• Downgrade Core, Julia 1.10 (lts) with the =floor-rewritten compat: telegraph 17/17, feedbackloop 12/12, birthdeath2D 18/18, tests passed. (CI: Downgrade (Core) is now green.)
• Core, Julia 1.12 (1): telegraph 17/17, feedbackloop 12/12, birthdeath2D 18/18, tests passed.
• Aqua test_ambiguities / test_unbound_args (1/1) / test_deps_compat (4/4), and JET test_package(...; target_defined_modules=true) (1/1): all pass on Julia 1.12 against the package project.
• Runic-formatted all changed .jl files (runic --check clean).

Remaining QA failure (upstream Catalyst — not fixable here)

The Aqua test_undefined_exports sub-check is still red. The undefined exports — CartesianGridReJ, infimum, iscall, params, supremum — are all inherited via @reexport using Catalyst; FiniteStateProjection's own exports (FSPSystem, DefaultIndexHandler, SteadyState) are all defined.

This is conclusively an upstream Catalyst bug, not an FSP defect. Catalyst itself fails the same Aqua check: running the equivalent of Aqua.test_undefined_exports against Catalyst directly reports [:CartesianGridReJ, :infimum, :params, :supremum] as undefined exports of Catalyst. FSP inherits those four plus iscall (the extra one is a @reexport double-hop: iscall is using-imported into Catalyst from TermInterface, so it is isdefined(Catalyst, :iscall) but is not re-bound in FSP).

• CartesianGridReJ (uppercase J) is a literal export typo in Catalyst — src/Catalyst.jl exports CartesianGridReJ while the type defined in JumpProcesses (and used everywhere else in Catalyst) is CartesianGridRej (lowercase j). The uppercase-J symbol is defined in no module in the dependency tree, so it cannot be made isdefined in FSP without inventing a fake binding. The typo is present on Catalyst 15.0.11 (the latest 15.x) and on Catalyst master.
• infimum, supremum, params are using-imported-then-exported by Catalyst from ModelingToolkit / Symbolics / DomainSets and do not propagate transitive bindings through @reexport using Catalyst.

Bumping to Catalyst 16.2 does not fix this (it yields a different set of undefined re-exports and additionally breaks FSP's Core tests — see the dependabot Catalyst-16.2 PR, whose Core jobs are red). Making this check green requires an upstream Catalyst fix (at minimum, fixing the CartesianGridReJ -> CartesianGridRej export typo). It cannot be fixed inside FiniteStateProjection without disabling/excluding the check or fabricating a binding.

Note on the Core (julia pre) CI job

On a later CI run this PR shows 3 failing asserts in birthdeath2D.jl (steady-state marginal/permutation tests, atol=1e-4) on the julia pre channel (1.13.0-rc1). This is not a regression from this PR: the same job was green on main, the resolved dependency set is byte-identical between the two runs (same SteadyStateDiffEq 2.11.0, NonlinearSolve 4.16.0, Catalyst 15.0.11, SciMLBase 2.153.1), and locally on 1.13.0-rc1 with that exact resolved environment birthdeath2D passes 18/18 (reproduced twice). The failures are borderline-tolerance numerical drift in the SSRootfind steady-state solve on the Julia release candidate, not a deterministic FSP change.

This PR makes Downgrade fully green and reduces QA from 5 failing Aqua/JET sub-checks to the single upstream-blocked test_undefined_exports check.

Please ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code