When reusing a dataset with metadata, item_count is incremented after being loaded from the metadata file. It leads to non continuous file increments, and breaks multiple functions on Datasets (export to file, etc).

Issue: code paths with/without metadata overlap in create_dataset_from_directory