QA: run_qa v1.6 form + ExplicitImports

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas.

Converts the hand-rolled test/qa/qa.jl (eight individual Aqua.test_* calls) to SciMLTesting 1.6.0's run_qa, with ExplicitImports enabled (explicit_imports = true).

qa.jl

using SciMLTesting, PolyChaos, Test

run_qa(
    PolyChaos; explicit_imports = true,
    aqua_kwargs = (; ambiguities = (; recursive = false))
)

• The eight Aqua.test_* calls collapse into Aqua.test_all via run_qa.
• The prior test_ambiguities(PolyChaos, recursive = false) tweak is preserved via aqua_kwargs = (; ambiguities = (; recursive = false)).
• No JET (the prior QA did not run JET).

ExplicitImports findings — all FIXED in src (0 ignore, 0 broken)

• no_stale_explicit_imports: dropped unused LinearAlgebra imports SymTridiagonal, eigen, pinv (verified unused in src/; not exported, not method-extension placeholders). Kept I, dot, issymmetric (used).
• no_implicit_imports: made the implicit using SpecialFunctions, SparseArrays, Distributions explicit:

  using SpecialFunctions: beta, gamma
  using SparseArrays: SparseVector, spzeros
  using Distributions: Beta, Continuous, Distribution, Gamma, Logistic, Normal,
                       Uniform, Univariate

• all_explicit_imports_via_owners, all_qualified_accesses_via_owners, all_qualified_accesses_are_public, all_explicit_imports_are_public: already passing.

Deps

• test/qa/Project.toml: SciMLTesting compat -> "1.6"; dropped SafeTestsets (unused by qa.jl; the harness provides @safetestset at the runtests level); Aqua kept (the ambiguities sub-check spawns a child process that needs Aqua as a direct dep). ExplicitImports not added (transitive via SciMLTesting).
• Root Project.toml: dropped Aqua from [compat]/[extras]/[targets].test (now exclusively a QA sub-env dep since QA runs in its own folder env); bumped SciMLTesting compat to "1.6".

Verification (local, released SciMLTesting 1.6.0, Julia 1.12.5)

QA group: 17/17 Pass, 0 Fail/Error/Broken (11 Aqua sub-checks + 6 ExplicitImports checks). All six EI checks confirmed passing individually.

Note: a separate pre-existing flaky Core test (test/polynomial_chaos.jl:40, "Mean and variance of beta distribution") can intermittently miss its atol = 1e-2 Monte-Carlo tolerance. Stress-tested on both clean master and this branch (~22k sampling checks each): identical behavior, maxdiff ~0.009 right at the tolerance edge. It is unrelated to this change (the explicit-import refactor produces bit-identical function bindings) and is not touched here; it is a master-level flake driven by unseeded randomized test inputs (α = rand():2:10).

🤖 Generated with Claude Code