Update performance plots to work after pipeline object is destroyed

[paulsonak]

Currently, the performance plots like perf_plots.plot_pred_vs_actual() only work on ModelPipeline objects when it would be useful to have a function that can recreate this plot from a model selected as the best model. I think the only one that couldn't currently be adapted is perf_plots.plot_perf_vs_epoch() since we don't save metrics from each epoch (although we did discuss doing that at one point..).

Request:
perf_plots.plot_pred_vs_actual_from_file()
perf_plots.plot_ROC_curve_from_file()
perf_plots.plot_prec_recall_curve_from_file()