Use i64 coordinate differences and checked i32 pitches in both scans and full reads. Reject nonpositive counts, malformed coordinate records, and pitches that cannot fit the packed schema.
Reuse library builders and cell traversal for full reads, lightweight hierarchy scans, and exact cell byte ranges. Preserve file offsets across readers and reject malformed ranges and FFI pointer/count inputs without aborting.
Document scan schemas and ID assignment. Add regression coverage for Arrow round trips, transforms, subset ordering, empty selections, and invalid input.