Looks interesting, but few comments on the forum & even a negative vote count ATM. Format kinda looks "old school" in terms of defining records, but I guess that can be a positive in some circumstances?
Modern data sources increasingly lean towards and produce nested and deeply nested semi-structured datasets (i.e. JSON) that are heavily denormalised and rely on organisation-wide entity ID's rather than system-generated referential integrity ID's (PK and FK ID's). That is a reason why modern data warehouse products (e.g. Redshift) have added extensive support for the nested data processing – because it neither makes sense to flatten/un-nest the nested data nor is it easy to do anyway.
If I were you though, I'd consider if I'd get more traction with an open source extension of Iceberg format that supports row based reporting and indexes for a unified open source HTAP ecosystem.
Gonna have a look at your DSL.