Auto-crop QA Validation Runner screenshots to the active window

The QA Validation Runner's Screenshot button (used throughout the manual-chapter screenshot workflow) currently saves a full-desktop PNG. The screenshot capture recipe's crop step then requires manually reading pixel bounds off the raw image and writing a one-off Python/Pillow crop invocation per screenshot — every single capture, no reuse across them since each dialog sits at different coordinates. Have the runner (or a small compass-side helper it shells out to) query the window manager for the target/active window's geometry (e.g. via xdotool getactivewindow getwindowgeometry or wmctrl -l -G on Linux) and crop to that automatically before saving, eliminating the manual bounds-reading step entirely.

This recipe is the standard path for authoring every future manual chapter (dozens of entities still lack one), and the crop step is pure boilerplate that differs only in the numbers, not the logic — exactly the kind of repeated manual work that should be automated once and reused, rather than re-derived by eye each time. It doesn't block chapter authoring today (the manual crop works fine), but it is friction that compounds across every remaining chapter.

• How do I capture screenshots for a manual chapter? — step 4, the manual crop this would replace.
• QA Validation Runner: in-app test tracking for the multi-worktree fleet — owns the Screenshot button this would extend.

• How do I document an entity? — the wider entity-documentation recipe this crop step is part of.