Fix Aqua QA failures: unbound type parameters in validate_cache and missing Pkg compat

[ChrisRackauckas-Claude]

Important

This PR should be ignored until reviewed by @ChrisRackauckas.

Fixes the failing QA test group on main (scheduled run 27210697854, QA julia lts/1 on ubuntu + macos).

Failures

Reproduced locally on Julia 1.10.11 (lts) with Aqua v0.8.16 via GROUP=QA (test/qa):

1. Unbound type parameters (Aqua.test_unbound_args):

   [1] validate_cache(::gType, ::Tuple{Vararg{ID, N_in}}, ::AbstractArray) where {gType, N_in, ID} @ src/interpolation_utils.jl:69
   [2] validate_cache(::DataInterpolationsND.EmptyCache, ::Tuple{Vararg{ID, N_in}}, ::AbstractArray) where {N_in, ID} @ src/interpolation_utils.jl:54
   

   In ::NTuple{N_in, ID} where {N_in, ID} the parameter ID is unbound when N_in == 0 (an empty tuple matches without binding ID).

2. deps_compat extras (Aqua.test_deps_compat): Pkg is in [extras] (used by test/runtests.jl to activate the qa/gpu subenvironments) but has no [compat] entry.

Root cause / why this "started" failing now

These failures are latent, not new. I verified that Aqua v0.8.14 (Aug 2025) flags the exact same unbound methods, and the missing Pkg compat dates to April 2025. They never surfaced because the QA matrix jobs previously ran with an empty GROUP env var and therefore silently executed the Core group instead of the QA group — e.g. the "green" scheduled run 26825993512 (2026-06-02) shows GROUP: (empty) in its QA job logs and no Aqua test summary at all.

The canonical grouped-tests workflow adopted in #68 sets GROUP=QA correctly on Linux/macOS, which exposed the failures on the next scheduled run.

Why Windows still "passes": it doesn't actually run QA. In SciML/.github tests.yml the step run: echo "${{ inputs.group-env-name }}=${{ inputs.group }}" >> "$GITHUB_ENV" has no shell: bash, so on Windows it runs under pwsh where "$GITHUB_ENV" (bash syntax) doesn't expand — GROUP is never set and the Windows "QA" jobs run the Core tests (their logs show Interpolations/Derivatives/DataInterpolations/Interface summaries, no Aqua). That workflow bug is reported separately in SciML/.github; it is not addressed here.

Changes

• src/interpolation_utils.jl: drop the unbound-parameter signatures. validate_cache(::EmptyCache, ::Tuple, ::AbstractArray) = nothing, and the fallback takes (cache, interp_dims::Tuple, ::AbstractArray) and builds the same error message via eltype(interp_dims) / typeof(cache). No dispatch behavior change (the NURBSWeights method remains more specific; Aqua's ambiguity check passes).
• Project.toml: add Pkg = "1" to [compat].

The Symbolics extension is untouched (under separate investigation).

Local verification (Julia 1.10.11 lts, linux)

QA group, Aqua v0.8.16 (also verified green with v0.8.14 pinned):

Test Summary: | Pass  Total   Time
Aqua          |   11     11  19.2s
Test Summary:   | Pass  Total  Time
ExplicitImports |    2      2  1.4s
Test Summary: | Pass  Total  Time
JET           |   12     12  5.6s

Core group (GROUP=Core Pkg.test()):

Interpolations |  206    206
Derivatives    |  369    369
DataInterpolations |  28   28
Interface      |   21     21
     Testing DataInterpolationsND tests passed

Runic check on the changed file passes (exit 0).

🤖 Generated with Claude Code