Transcode from other formats besides CSV

[jlowe]

When generating NDS data sometimes we want to generate for multiple formats. For example, we might generate Parquet data and then also want ORC data. It's more efficient to transcode from one columnar format to another rather than only transcode from the raw CSV. It also is handy when you want to generate ORC data but only the Parquet transcode output is available and the raw CSV has already been deleted.

It would be nice if nds_transcode.py supported options to specify the input format to use rather than assume CSV. Parquet/ORC would be helpful, but for bonus points it would be nice if all of the output formats were also supported as input format possibilities.