CouchDB backup restoration: decide between _replicate endpoint and direct file copy for large databases

The task is to evaluate which CouchDB backup restoration approach—using the _replicate endpoint or performing a direct file copy—preserves low‑level database metadata such as compaction state and attachment store format while avoiding data races during the restore.

Uncertainty remains whether _replicate captures the full file system metadata required for large databases, and whether a direct file copy can be executed without exclusive access to the source database. Specific questions: Does _replicate retain the database file’s compaction state and attachment store compatibility? Can a direct file copy be performed safely while the source database remains online? Which method guarantees consistent attachment checksum verification after restoration?