feat(prediction): AAPredPlot.eval bars-with-hue + full AAPred notebook param coverage + CI gate

[breimanntools]

Three coupled changes, all off master, 335 prediction + api-meta tests green.

1. AAPredPlot.eval → grouped bars with hue = model (method comparison)

eval compares methods (models × metrics), so it is a grouped bar plot — one colored bar per model (the hue), grouped by metric, with cross-validation score_std error bars, hatched held-out bars, and an optional chance baseline. Comparing CPP parameter combinations is a different job, already covered by the feature-optimization protocol aap.find_features + its aap.plot_eval heatmap, which eval's docstring now points to. (No kind/cmap; no heatmap branch.)

This also settles the intent behind #310 — the "evaluation heatmap" is the parameter-range heatmap (aap.plot_eval), while eval is the method-comparison bar plot.

2. Full parameter coverage in all 14 AAPred/AAPredPlot example notebooks

They previously demonstrated 1–2 params each; every one now demonstrates every public parameter of its method (as explicit keywords, grouped), re-executed with saved outputs.

3. A CI gate so this can't regress

• tests/unit/api_tests/test_notebook_param_coverage.py: every method with an example notebook must demonstrate every public param (name-based). Prediction notebooks are held to zero gaps; a committed notebook_param_coverage_baseline.txt ratchets down the pre-existing ~94-method / 257-param backlog (pyright-style — may only shrink, with a staleness guard forcing burn-down).
• .claude/rules/notebooks.md upgraded from an aspirational note to a test-enforced rule.

Root cause of the recurring gap: the notebook checker existed as a skill script but was never wired into CI or run. It is now a real gate. Backlog burn-down tracked in a follow-up issue.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.83%. Comparing base (7dcc8d8) to head (17b7a1a).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #344   +/-   ##
=======================================
  Coverage   94.83%   94.83%           
=======================================
  Files         196      196           
  Lines       18767    18767           
  Branches     3175     3175           
=======================================
  Hits        17797    17797           
  Misses        633      633           
  Partials      337      337           

Files with missing lines	Coverage Δ
aaanalysis/prediction/_aa_pred_plot.py	94.70% <100.00%> (ø)

Components	Coverage Δ
cpp_core	94.95% <ø> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.