- Move memtable node creation out of lock and reduce allocations in neighbor connections for better performance
- Optimize vector handling by sending merge vectors only when needed and skip vector transfer for searches unless requested
- Backup now splits large shards into smaller chunks and cleans up unused backup code