Core DataDrivenDataset tests fail on exact floating-point equality

[ChrisRackauckas-Claude]

On current origin/master (b9fd6fb6, PR #627), the top-level Core test group fails in test/Core/problem.jl on exact equality checks in the DataDrivenDataset testset.

Reproduction from a clean checkout of SciML/DataDrivenDiffEq.jl master:

GROUP=Core timeout 3600 julia --project -e 'using Pkg; Pkg.test()'

Observed failure:

DataDrivenDataset: Test Failed at test/Core/problem.jl:159
  Expression: b(s2) == b(s1)
   Evaluated: [... -0.9589242746631385; ... 0.28366218546322625] == [... -0.9589242746631386; ... 0.2836621854632259]

DataDrivenDataset: Test Failed at test/Core/problem.jl:162
  Expression: hcat(X, X, X) == b(s5)
   Evaluated: [... -0.9589242746631385; ... 0.28366218546322625] == [... -0.9589242746631386; ... 0.2836621854632259]

Test Summary:                                | Pass  Fail  Total
DataDrivenProblem                            |   75     2     77

This looks like benign floating-point roundoff from recomputing the same trigonometric data through different dataset/basis paths, but the test currently uses exact ==. CI shows the same failure across lts, Julia 1, and pre-release jobs.

This was noticed while checking unrelated PR #631; the failure reproduces on unmodified origin/master without that PR's changes.

[ChrisRackauckas-Claude]

Draft fix PR opened: #634. It changes the exact dataset comparisons to approximate comparisons and locally passes GROUP=Core.