We now require CoreNLP listening on port :9000. We will have to find
a way to spin that up in CI as well.

This was removed from `dev` and was causing the DB tests to fail.

IF NOT EXISTS makes sense only when you provide the name of the index.

Unique index is necessary when one wants to refresh the view
concurrently.

This is so that we correctly order _a_history later.

[*r*epoSaver*A cEtxicoen]p "rtepioosn/:no deSqlError Id{-0s-tqmlp-Srteatepo1 2=3 8"22-3150013."c,b osrq"l
E[wrixteecNSotdaetSutso r=i eFsa]t a[l(E)r]r
or, sqlErrorMsg = "une instruction insert ou update sur la table \194\171 context_node_ngrams \194\187 viole la contrainte de cl\195\169\n\195\169trang\195\168re \194\171 context_node_ngrams_context_id_fkey \194\187", sqlErrorDetail = "La cl\195\169 (context_id)=(135) n'est pas pr\195\169sente dans la table \194\171 nodes \194\187.", sqlErrorHint = ""}

Also, pin postgres to v11 so that we're compatible with Debian
upstream.

This caused subtle bugs when ALTER TABLE with search_title was
called. It is a wrong practice to define an index immutable when in
fact it's not. The count(id) part makes it mutable in fact.

This reverts commit 29ee1972.

It introduces a bug, it should not be merged.

However, the code doesn't work currently.

This has couple of drawbacks:
- we use postgres-simple and a raw sql query, opaleye would be better
I guess
- there is a hack for empty query, this should be replaced by
removing the WHERE clause in this case