feat(viz): per-facet shading for faceted regression-fit plots

[ismayc]

Why

Two faceted multiple-regression inference plots in Chapter 10 showed the correct bootstrap/null distributions but without the shading overlay — shade_p_value/shade_confidence_interval were scalar-only and couldn't shade per facet. This adds array/per-term shading so those figures shade like the R book, in both the plotly (default) and plotnine engines.

What

shade_p_value and shade_confidence_interval now accept per-term inputs:

• p-value: an observed FitResult (term + estimate), a term-keyed frame (term + estimate/stat), or a dict
• confidence interval: a per-term CI table (term, lower_ci, upper_ci)

Each facet of a visualize_fit() plot is then shaded from its own observed statistic / interval.

• ShadeSpec carries an optional per_term mapping; InferPlot remembers the facet terms from visualize_fit() and dispatches to per-facet shading.
• plotly: shades each subplot via add_vline/add_vrect with row/col (infinite tails clipped to that subplot's data range).
• plotnine: term-keyed geom_rect/geom_vline layers (inherit_aes=False) land in the correct panel.
• FitResult.visualize() and visualize_fit() gain shade_pvalue=/shade_ci= keyword args; + composition also works.

Usage:

obs  = specify(df, formula="y ~ a + b").fit()
null = specify(df, formula="y ~ a + b").hypothesize(null="independence").generate(reps=1000, type="permute").fit()
visualize_fit(null) + shade_p_value(obs_stat=obs, direction="two-sided")   # per-facet

Notes

• Scalar shading on single-panel plots is unchanged.
• Adding a per-term spec to a non-faceted plot raises a clear TypeError.
• 100% coverage retained (169 tests), ruff clean.

🤖 Generated with Claude Code