Vignette: default against default_tuned for bull trout in the Parsnip, side by side #342

Why

link's value is that a model is a config you can change and compare. The README figure should show that: two configs on one piece of watershed, side by side. It should come from a vignette that tells the story behind the difference.

Tasks

• Build both configs on one segmentation. Prepare PARS once at the current HEAD, then classify BT under default and under default_tuned on those same segments, using the variants harness (data-raw/habitat_variants_build.R). Two separate full runs would not do: a full run is not byte-reproducible (the PSCIS-to-modelled crossing tie, Per-bundle habitat thresholds CSV in config.yaml #282), so the maps could differ for reasons other than the tuning. The harness's guard stops at the current HEAD because default's thresholds have moved since default MAD ranges for BT, GR, KO and RB, from its own width minima #307; unblock it first.
• Pick the reach. Find the PARS sub-basin where the BT rearing gradient change (0.1049 → 0.1349) moves the most km, and choose one where the change reads clearly on a map.
• Write it up, as a section of the PARS vignette or a short vignette of its own: the threshold row that changed, side-by-side maps of the sub-basin, and the km difference. Data is cached through data-raw/, as the PARS vignette's is.
• Make the README figure from the vignette's cached data with data-raw/readme_figure.R. It replaces the single-config grayling map.

Relates to #338 and #284.