It's faster now, makes less joins, doesn't need VALUES.

This was (User, Either CorpusName, [CorpusId]) before, but the case of
UserMaster doesn't make sense with these parameters, so I rewrote the
function to accept only correct datatypes as inputs.

We had CoreNLP hardcoded, however with Spacy this resulted in
selectLems returning no entries, because the algo_id from DB didn't
match. Hence no singulars grouping.

This contains only data, removes some Database dependency in imports.

This reverts commit f4284b4e.

This paves the way to a more effective search in corpus by modifying the
internal "leaves" type of the `Query` type from a `Term` to a new type
called `QueryTerm`, which can either be an extact match or a partial
match, which can be rendered into a proper Postgres TS query via the
":*" syntax.

Also, some test refactoring and add servant-client to tests.

This commit refactors ever so slightly the '.Stem' modules to create a
unified interface; now Gargantext.Core.Text.Terms.Mono.Stem exports a
single 'stem' function which can be used with different
`StemmingAlgorithm`s, and now the other algorithms' functions have been
shielded behind internal modules.

This makes possible to see at glance which part of Garg are using which
stemming algorithm, as it wasn't always clearer before.

This also allows us to have a better control over the langugages, as not
all the stemming algorithms supports all the languages that Garg
supports.

Removing the `All` data constructor from  the `Lang` datatype has
several benifits, the main one being that we now have total functions
for things like `toISO693`.

This will make possible to write an uniform interface for stemming
`stem :: Lang -> StemmingAlgorithm -> T.Text -> T.Text`, as now
`Lang` will always map to a valid (ISO693) language.

This reverts commit efbe327f, reversing
changes made to 6d9bcd07.