Consider building around a parquet-based spec

[daniel-j-h]

Now that we have a bit of experience with our protobuf based home grown file format, let's consider a future evolution. In particular the wider ecosystem seems to converge around parquet to build custom formats on top e.g. see https://geoparquet.org

We should consider building something like GraphParquet as an open spec and have our experience from this project go into it. The parquet file format from what I can tell is more than capable to represent our graph datastructures and does some basic level of compression we can work with to make a compressed sparse row graph small when stored.

Evaluate this and come back with insights and thoughts.