feat: prediction-score histogram + plot_rank error bars/class colors

[breimanntools]

Part of #305.

Problem

Two prediction-result figures from the original project have no library API and
are reused across the paper but must be hand-built each time:

1. Prediction-score distribution — per-class histogram of a model's
   prediction score (0–100%) showing how substrate / non-substrate / unknown
   separate (plot_pred1 / _plot_pred1, scripts/plot_cpp_pred.py): a
   sns.histplot(hue=..., binrange=(0,100), multiple="stack") plus ~30 lines of
   int-ytick / xlim / despine / custom-legend fiddling.
2. Ranked candidates with agreement — proteins ranked by predicted score as
   horizontal bars colored by class, with per-protein std error bars and gene
   yticks (plot_pred3_top_hits, reused 3×): sns.barplot(orient="h") +
   plt.errorbar(fmt='none', capsize=...).

The library already has plot_rank, but it has no error-bar / class-color
variant for the "named top-N candidates" figure.

Goal

Add a prediction-score histogram and extend ranking so the two standard
prediction figures are one call each, driven by a model's per-sample scores.

Requirements

• aa.plot_prediction_hist(df_pred, score=..., group=..., ...) (class-
  separated score distribution); consider a TreeModelPlot/ShapModelPlot
  home.
• Extend plot_rank (or add a sibling) with per-item error bars + class
  colors + optional cutoff line, for the ranked-candidates figure.
• House colors/legend via existing helpers; plotting.md compliance.
• numpydoc + Examples include.

KPIs / Acceptance criteria

• Both figures reproducible in one call each from a df_pred with score /
  std / class columns (visual + shape check).
• plot_rank default output unchanged (additive args only; regression).
• ≥1 unit test per new symbol/arg (positive + negative).

Scope / non-goals

• Plotting only; the model panel that produces the scores is item-4 / Model evaluation & comparison: repeated-CV + bootstrap CIs + paired ΔMCC #91 /
  aap.predict. SHAP-driven scores are item-8.

Dependencies

• Part of epic epic: use-case convenience methods — collapse γ-secretase notebook boilerplate #305 · relates Visualization layer (multi-scale explanation) #56 (visualization layer).

Standards checklist

• CONFIRM-FIRST (__init__.py/__all__ if new symbols) · plotting.md
  compliance · numpydoc · tests · no-print

[breimanntools]

Delivered on master via #332: the prediction-score histogram is AAPredPlot.hist, and the ranking with std error bars + class colors is AAPredPlot.ranking (verified: errorbar/col_std + class colors). Consolidated onto AAPredPlot rather than extending the standalone plot_rank.