Evaluate using `Jina` for Blue Brain Search

[FrancescoCasalegno]

Jina has very nice features that could really simplify our code, especially when it comes to:

• computing cosine similarity for (query, item) pairs of vector embeddings (also at a scale, w/o having to load all vectors into memory)
• implementing servers with RESTful APIs for embeddings and for search.

For this issue, we should first try to have a PoC to see how well would Jina scales when dealing with large number of papers, assess if we would still need a sentences table in our DataBase, and more in general which parts of our workflow can leverage Jina.