Add tagging, multi-error, best-of-family, timeout, autodiff, and autoplot to DiffEqDevTools

[ChrisRackauckas-Claude]

Summary

Implements Phases 1-7 of SciML/DiffEqDevTools.jl#177 for the DiffEqDevTools sublibrary now in this monorepo.

• Phase 1 — Core tagging infrastructure: tags::Vector{Symbol} field on WorkPrecision, filter_by_tags, exclude_by_tags, get_tags, unique_tags, merge_wp_sets. Also fixes StructArrays 0.7 compat via _dicts_to_structarray.
• Phase 2 — Multi-error-mode: error_estimates kwarg on WorkPrecisionSet to compute multiple error metrics (:final, :l2, :L2) in one pass. available_errors() helper.
• Phase 3 — Reference method plotting: reference_tags and reference_style kwargs in plot recipe for distinct baseline method styling (dashed/thin/transparent). Tag-based filtering (tags, include_tags, exclude_tags) in plot recipe.
• Phase 4 — Best-of-family helpers: wp_area (log-log trapezoidal area), best_by_tag, best_of_families for automatic cross-family comparison.
• Phase 5 — Time cutoff: timeout kwarg on all WorkPrecision/WorkPrecisionSet constructors. Exceeding timeout marks results as NaN with a warning.
• Phase 6 — AutoDiff comparison helpers: with_autodiff_variants creates AD-variant setup dicts with proper tagging.
• Phase 7 — Autoplot: autoplot(wp_set; families, reference_tags, best_n) generates per-family subsets, best-of-families, and full set as Dict{String, WorkPrecisionSet}.

All changes are fully backward compatible — new fields have defaults, new parameters are keyword-only.

Test plan

• 63 new test assertions across 6 new test files all pass locally
• Existing ode_appxtrue_tests.jl passes (backward compatibility confirmed)
• CI validation

🤖 Generated with Claude Code