Disallow zero-len dataframes when parsing

[LucaMarconato]

Parsing dataframes with zero length may lead to subtle bugs and has probably no use case. An example is to to_zarr.py script for the cosmx_io data in the sandbox. Parsing a zero-len dask-dataframe of points lead to a problem with unknown partitions.

We should

• [ ] check for, and disallow, zero-len data in the parser.
• [ ] add tests
• [ ] also add a test to reproduce the bug with the cosmx data.