Generate reviewable evidence inventories instead of pinned counts

[woksin]

Adding a story, part or export no longer means hand-editing a count in a verifier. The verifiers compare against committed inventories of names instead, so a failure says exactly what was added or removed. Pinned counts broke CI three times.

Changed

• The Storybook, parts manifest, renderer, kernel and export verifiers compare the current state with committed, sorted inventories (Source/scripts/*-inventory.json, Storybook/scripts/*-inventory.json) instead of fixed counts. When something is added or removed, the failure names it and says how to update the inventory. (Generate reviewable evidence inventories instead of hardcoded counts #217)
• Run yarn generate-inventories to refresh every inventory after a deliberate change, then review the diff like any other change. CI only compares; it never regenerates. (Generate reviewable evidence inventories instead of hardcoded counts #217)
• The renderer inventory records each renderer's reviewed display name as well as its ID, so the composed Storybook manager still fails when a renderer label changes. (Generate reviewable evidence inventories instead of hardcoded counts #217)
• The checks that every renderer has the same stories, and that the built-in preview comes first, are unchanged. (Generate reviewable evidence inventories instead of hardcoded counts #217)

Closes #217.