Directory structure & metadata for parquet files

[mjcollin]

Metadata is in #7 but if it's going to be in a separate file, then that needs to be considered here too.

The nested or faceted style of storage is really cool ("/guoda/data/gbif-idigbio.parquet/source=gbif") and intuitive to use. However coding mistakes like the example just there (it doesn't have the date on the end so it loads two copies of GBIF) can be exponentially disastrous as we add more sources and versions of data sets. Since we're including targeting intermediate programmers with this platform, do we want a safer syntax?

[mjcollin]

A possible compromise: The archive version of data storage uses Jorrit's pattern and the latest is a copy of the latest at a fixed file name in the existing data folder - HDFS does not support symlinks so we'll have to have copes somewhere.