Add GPUArraysCore extension: broadcast-based seed! for GPU dual arrays

[ChrisRackauckas-Claude]

Summary

ForwardDiff 1.x rewrote the four seed! methods in src/apiutils.jl to write each
dual with a scalar setindex! loop over structural_eachindex:

for (i, idx) in zip(1:N, structural_eachindex(duals, x))
    duals[idx] = Dual{T,V,N}(x[idx], seeds[i])
end

On GPU arrays this triggers ERROR: Scalar indexing is disallowed, so every
ForwardDiff.jacobian / jacobian! call on a GPU array (e.g. CuArray) errors
in seed!. ForwardDiff 0.10 seeded with broadcast and worked on GPU arrays, so
this is a regression for GPU usage introduced in the 1.0 rewrite.

This was first hit downstream in SciML/ComplementaritySolve.jl#65, where the GPU
solver paths (NonlinearSolve's AutoForwardDiff jacobians) broke once the
environment resolved ForwardDiff ≥ 1.

Fix

Add a GPUArraysCore package extension (ForwardDiffGPUArraysCoreExt) that
overrides the four seed! methods for AbstractGPUArray{<:Dual} with
broadcast-based seeding, restoring the pre-1.0 behavior. GPU arrays are always
dense, one-based, and carry isbits element types, so the
structural_eachindex / unset-element handling of the generic methods is
unnecessary on this path and plain broadcast suffices.

This is a weak dependency, so it adds nothing for users who don't load a GPU
array package.

Scope

This covers the jacobian paths (vector and chunk mode), which is what seed!
feeds. ForwardDiff.gradient on GPU arrays has a second scalar-indexing site in
the gradient extraction path (extract_gradient_chunk!) that this PR does not
touch; fixing seed! is the necessary first step and resolves the jacobian
regression.

Tests

test/GPUArraysCoreTest.jl exercises all four seed! methods through
jacobian, jacobian!, and chunked configs using JLArrays, which emulates
GPU array semantics (including the scalar-indexing ban) on the CPU — so the
extension is covered in CI without a physical GPU.

---

Note

Opened as a draft by an agent on behalf of @ChrisRackauckas. Please ignore
until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code

[ChrisRackauckas]

Because the broadcasts were purposefully removed due to performance things, we should at least recover the GPU functionality by handling AbstractGPUArray.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.88%. Comparing base (777420f) to head (18a692a).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #816      +/-   ##
==========================================
+ Coverage   90.74%   90.88%   +0.13%     
==========================================
  Files          11       12       +1     
  Lines        1070     1086      +16     
==========================================
+ Hits          971      987      +16     
  Misses         99       99              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.