Fix downgrade bounds and dataset comparisons

[ChrisRackauckas-Claude]

This PR should be ignored until reviewed by @ChrisRackauckas.

Summary

• Replace exact floating-point equality checks in DataDrivenDataset tests with approximate comparisons where equivalent symbolic basis evaluations differ by roundoff.
• Raise stale lower bounds that made downgrade CI produce incompatible resolver graphs for root, DataDrivenSparse, DataDrivenSR, and DataDrivenLux.
• Add test-only DataStructures bounds for sublibraries whose downgraded test manifests otherwise pin DataStructures too low for SymbolicUtils/MultivariatePolynomials.

Root Cause

The current Symbolics/ModelingToolkit/SciML stack requires newer transitive versions than several sublibrary lower bounds allowed. The downgrade workflow minimized direct test dependencies into combinations that Pkg could not resolve during Pkg.test. The Core test failure was separate: equivalent dataset constructors evaluated through slightly different floating-point paths, so exact equality was too strict.

Validation

• GROUP=Core timeout 3600 ~/.juliaup/bin/julia +1.10 --project=. -e 'using Pkg; Pkg.test()'\n- DataDrivenSparse downgrade action reproduction + Pkg.test(... allow_reresolve=false)\n- DataDrivenSR downgrade action reproduction + Pkg.test(... allow_reresolve=false)\n- DataDrivenLux downgrade action reproduction + Pkg.test(... allow_reresolve=false)\n- timeout 3600 julia --project=.runic-env -m Runic --check src test docs lib\n