Currently our code looks like this:

https://gitlab.iscpif.fr/gargantext/haskell-gargantext/blob/79490deb7d2e2459d79620488e06b61d5c2bd43a/src/Gargantext/Database/Action/Flow.hs#L279

We call flowCorpusUser which makes the CoreNLP query after all docs are imported. However, the external APIs are fragile and with longer queries they fail often. It is bad that ngrams are empty after such import. We could rewrite this so that CoreNLP is called alongside of API fetch, in conduit. Later we could parallelize that even further so that the API fetch and CoreNLP are called in parallel.