Classifies all 18,439 vol1 MIT Press CE annotations into 16 buckets
(bulk-* rules, queries, layout, micro-edit, etc.) and tags each with
needs_locator. Single-character silent strikes/inserts are bucketed as
'bulk-microedit' (covered by the global bulk passes) so the Phase B
locator focuses on the ~4k substantive entries that actually need to be
diff'd against the QMD source.
Output: ledgers/vol1-annotations-ground-truth.json plus per-chapter
summary.
