- metadata.py: persistent ExifTool session (avoids Perl startup per file),
filters out File:/ExifTool: noise
- scanner now populates observations.meta as JSONB
- size duplicated into observations for self-contained queries and to
strengthen the rescan idempotency check (path + mtime + size)
- README rewritten with state diagram, schema tables, scan/apply workflow
- archive.py: CAS path layout (blobs/<2>/<2>/<hash>), view paths
scoped by source_kind, hardlink helpers
- applier.py: cursor-paginated apply_pending with dry-run support
- mama-apply CLI with progress and --dry-run
Note: cross-dataset hardlinks fall back to copy (POSIX limitation),
so applying from /mnt/preview to /mnt/archive currently doubles
storage. To be addressed by consolidating sources into the archive
dataset or by introducing --delete-source for move semantics.
- observations now keyed by (path, mtime) for idempotent rescans
- new index ix_observations_path_mtime
- mama-dev reset: truncate all data, schema kept
- mama-dev stats: overview, breakdowns by source_kind/status/hostname/MIME, largest blobs
