What about hdf5? I'm starting to use it as an in-memory, parsed cache representation for data stored in like a plaintext file document database. With python's PyTables and visidata acts as the independent data explorer app.
I like HDF5 because it's compatible with everything, but it's a bit more difficult to get started with than other modern formats. Partly this is due to design by committee including everything, and the APIs follow suit. With some more modern APIs I think it could come back as an archival format.
Hdf5 files are good if the data is write once, and numeric unless things have changed over the last fives years. From my recollection, you can’t append data so you have to rewrite the whole table and text fields are fixed width.