rather than three KVs with the same keys, what’s the memory usage if you use a Polars DataFrame or an Arrow RecordBatch where each u64 key is only written exact

rather than three KVs with the same keys, what’s the memory usage if you use a Polars DataFrame or an Arrow RecordBatch where each u64 key is only written exactly once? How does that impact the lookup times?