Fix plot idxs dimension check to use transform output arity - #660

Summary

Same fix as SciMLBase#1609 for the copy of the plot helpers that SciMLBase PR #1305 intends to delegate to: stop asserting length(var) - 1 == dims on idxs input arity; check each transform's output coordinate count and throw ArgumentError when series genuinely disagree (2-D vs 3-D).

Follow-up (independent review #1): (1) short-label branch now also requires x[1] === DEFAULT_PLOT_FUNC so (adder,0,3,4) stays "f(t,u[3],u[4])". (2) deleted probe helpers; dims come from evaluated map(f, ...) output (avoids MethodError/DomainError on transforms that used to plot).

Follow-up (independent review #2): restored the exported solplot_vecs_and_labels(dims, vars, plott, A) 4-arg signature (dims accepted and ignored; computed output dims count); deleted the orphan history comment; added a test calling the 4-arg form.

Part of SciML/DifferentialEquations.jl#360 — primary user-facing fix is SciML/SciMLBase.jl#1609.

Review follow-up: before (new tests on prior PR code)

Expression: labels == ["u[3]", "f(t,u[3],u[4])"]
 Evaluated: ["u[3]", "u[4]"] == ["u[3]", "f(t,u[3],u[4])"]

MethodError: no method matching (::var"#g#g##0")(::Int64, ::Int64)
  @ RecursiveArrayTools plot_series_output_dims ...
plot idxs ... (#360) |   13     2      1     16

Review follow-up: after

plot idxs with mixed input arity but matching output dims (#360) |   27     27  3.7s
Public API Tests |   49     49  (in GROUP=Core)

Test group tail (GROUP=Core via Pkg.test)

Test Summary:        | Pass  Total   Time
Linear Algebra Tests |   54     54  22.5s
Test Summary:         | Pass  Total  Time
SciMLStructures Tests |   53     53  6.5s
Test Summary:     | Pass  Total  Time
Measurement Tests |    2      2  0.1s
     Testing RecursiveArrayTools tests passed

Not verified

• SciMLBase#1305 merge / that SciMLBase actually calls this RAT path after delegation.
• Plots.jl end-to-end rendering of the CoupledSprings MWE.

What to push back on

• Whether this should land before or after SciMLBase#1305 merges.
• custom transforms (f,0,i) are now labelled f(t,u[i]) instead of u[i]; this is more accurate (e.g. (sq,0,3) plots u3²), but it is a visible label change.

Please ignore this draft until reviewed by @ChrisRackauckas.

Risk assessment

• Risk: low
• Blast radius: the plot helper in src/vector_of_array.jl (DiffEqArray/VectorOfArray recipes). The exported solplot_vecs_and_labels(dims, vars, plott, A) keeps its 4-argument signature and ignores dims. The logic is identical to the merged Fix plot idxs dimension check to use transform output arity SciMLBase.jl#1609. Visible change: custom transforms (f,0,i) are labelled f(t,u[i]) instead of u[i].
• Evidence: the new tests fail before and pass after (rat | 49 49). Plotted data for existing forms is unchanged. The two red CI jobs (SciMLSensitivity Downstream Core1 and Core6) fail the same way on master: https://github.com/SciML/RecursiveArrayTools.jl/actions/runs/36085624927.
• Independent review: Claude Code (Opus 5.5), three rounds. The final one, on 4df9547 (the current head), rated it low with high confidence, verdict MERGE. The one remaining nit is optional.
• Merge: auto-merge candidate. It keeps RAT in step with the merged SciMLBase#1609.

Risk assessment and review coordination by 🤖 Claude Code (model: claude-opus-5-5[1m]), https://claude.ai/code/session_01NdFVhANTF4nepTbqYxjfRP