Benchmark new Datumaro structures #1941
Description
This ticket is meant to ensure datumaro functions properly with large datasets. Consider that very large datasets might need to be sharded and check that the performance doesn't degrade in large datasets.