Add --metadata-only mode; harden resume and idempotency

- Add --metadata-only flag: fetches scan detail pages, writes
  metadata.json + scans.csv rows, skips all image downloads.
  Re-runs skip scans whose metadata.json already exists.
- Atomic progress.json saves (temp-file rename).
- Heal-on-resume: tiles on disk but not in progress are silently
  re-marked before building the pending list.
- scans.csv dedup: skip row if mosaic URL already in progress.
- Rename mosaic_downloaded -> mosaic_on_disk (reflects disk state).
- --recheck now checks mosaics as well as tiles.
- RunStats dataclass replaces raw int return; richer run summary.
- Fix argparse allow_abbrev reverted; fix --scan-id + --metadata-only
  glob fallback when scan_time is absent.
- Add .venv/ to .gitignore.
- README: fix typo, update worker counts, document all new behaviour.

spruce/session.py

@@ -263,10 +263,6 @@ class MachineSession:
         }
         if dry_run:
             return row
-        if dest.exists():
-            row["downloaded_at"] = "already_exists"
-            row["file_size_bytes"] = dest.stat().st_size
-            return row
         size = self.download_file(tile["url"], dest)
         if size:
             row["downloaded_at"] = datetime.now(timezone.utc).isoformat()