ENH/REF plot_partial_residuals versus plot_partial in GLM/GAM

[josef-pkt]

Based on quick skimming, GLMResults.plot_partial_residual is just a CPR (component plus residual) plot for a single column of exog.
However it uses the weights, weighted residuals in the plot. (those might be one of the available residuals)

GLMGamResults.plot_partial plot partial prediction and optional cpr scatter points but only for the linear prediction, using resid_working.
It also adds confidence intervals for the partial linear prediction.

regression plot for OLS also have a CPR plot for a single exog column.

what can we improve, enhance or make more consistent?

• prediction intervals ? ENH: diagnostic plot partial predict confidence interval #2621
• dependence on multicolumn exog (e.g. splines, polynomial terms)
• what happens with categorical exog?
• which residuals
• how generic can we make this

(I found plot_partial_residual by tab completion because it was next to the new GAM plot_partial)