Fix for the bug encountered by Stephan +
rewriting of the whole building of the
bioagressors' table (with consequences on the
alvisNLP plans for the annotation).

If the date still contains YYYY or MM, we try to
look for the missing info the metadata of the
bulletin. The original, purely text-based
normalisation is still available under
nav:features:ISO{0} (could be used in the rdf
export to attach a prediction quality).

Preferrable to do this early so I don't have to
keep repercuting this change in the various exports.

Ajout d'une version de `src/alvisNLP/utils/query_endpoints.py` compatible avec l'hébergement de FCU, PPDO et des BSV sur le serveur openstack du CATI CODEX.

Adaptation des définitions du dataset en vue d'une migration vers opendata.inrae.fr/bsv-def et opendata.inrae.fr/bsv-res.

The list of manual bioagressors is integrated, with all that it entails
(lemmatizing vernacular names, looking for entity types...).
It's not pretty but it works.

Document `./2014/ProgrammeLaboVert_cle058f1c.html` has been removed as
it's not part of the vitials corpus.
I don't know why it was there in the first place, perhaps in an older
version of the triplestore it was put there ?

- Sorted the entries
- Added some entries that introduce noise in the annotation

Removing mentions inside of other mentions
E.g "phytoplasma solani" is removed when "candidatus phytoplasma solani"
is detected

Case folding is acceptable on every token's first caracter
e.g Candidatus Phytoplasma Solani -> Candidatus phytoplasma solani
is detected.

For timespans of the form "du 12 au 15 mai":
- new normalisation in @predicted-ISO where the first month
corresponds to the second month

- manual removal of a label
- code is exported

- compound-dates of the form \d\d/\d\d could be interpreted
as both month-year and day-month, day-month is preferred.
- dates containing \n or \t are removed (e.g JUIN\n25)
- dates containing both \ and a whitespace are removed (04/08 19)

Entities are represented as their own nodes.
Link between the actual tokens and the entity nodes are encoded
as semantic relations with label `entity`.

Also remove the duplicates between NCBI and taxref
prioritising the taxref one.
We could check that the norms are equivalent but don't do it
currently.

- Noisy "maturity" label is dealt with
- Noisy Baggiolini codes only kept when they're unambiguously a stage

- Check the normalisation
- Add to the entities layer with appropriate features for rdf export

Resources are nammed according to the patterns
we defined together.

With one exception :
we don't put the labels of the normalisation in the URI.