Since we represent pose tracking data as xarray objects, it could be attractive to save them to zarr, especially since xarray already provides a xarray.Dataset.to_zarr() method.
This would allow for chunking and parallel read/write operations.
This could become the movement-native way of saving/loading pose tracks.

That said, we should coordinate with the animal pose estimation community at large to hopefully converge on a common file format for such data, as mentioned in this paper