Improve taxonomy parser performance by serializing writes

[alexgarel]

The parser in parser/openfoodfacts_taxonomy_parser/parser.py create_nodes, creates nodes one by one, whereas it never makes a request to deduce what to create.

So instead of creating data one by one we should:

• create a list of all the queries we need to run
• batch run them

The same applies to create_child_link, where requests could be run in batch.

This might improve performance by a important factor.

One elegant way to do this kind of batch request is https://neo4j.com/docs/cypher-manual/current/clauses/unwind/, but I'm not sure we need it.