We have use cases that require multiple versions of the same datasets to be available. For example, we have a knowledge graph made of entities of relations, and we have multiple versions of the knowledge graph that's distinguished by schema names right now:

knowledge_graph__1_3 (version 1.3)
    - entities
    - relations
knowledge_graph__1_4 (version 1.4)
    - entities
    - relations
knowledge_graph__2_1 (version 2.1)
    - entities
    - relations

While we can shoehorn the use cases by using names to version datasets, it doesn't feel like an elegant way. I'm also aware of the Delta Lake versioning capabilities, but we would like to keep each version a top-level artifact rather than historical versions as in the Delta Lake case. Any recommendations of best practices?