Add parquet format as an output for embeddings?

[hlapp]

The Embedding Explorer (and image search?) uses a Parquet database for input of vision embeddings. Maybe we should offer the option here to generate Parquet right here to reduce friction from getting from pybioclip output to Embedding Explorer input.

@NetZissou thoughts? Is there a spec for required columns?

[egrace479]

There are only two required columns, but there are nice to haves based on TOL-200M data, and unenumerated columns can be read as well.

Including a parquet output option sounds like a good idea to me.