Increase granularity of "precision and recall by threshold" plot

This plot's smoothness hides interesting and useful performance issues, e.g. a flat part of the recall curve helps identify cases that have similar scores and are all wrong, and we might find ways to improve model performance by taking a closer look at them. The current plot skips from 1% to 5% to 10% to.... If there were a flat part of the recall curve, it would be hard to tell.

Given that we're already pulling the data and making the sklearn calls, this plot should show more granularity (in most applied cases, the curve will be jagged)