Switch to MongoDB

[SeguinBe]

For now, the DataManager is very dumb. It does not save the data, and needs to recreate the whole search matrix every time an image is added to the system.

Because the searching data might get too big in the future to be stored in memory, we need to move to a cleverer approach.

Proposal :

• All the information stored in MongoDB
• Search matrix (which is the normalized concatenation of the signatures of the elements) is recomputed periodically
• Search matrix may be a memmap version of a numpy array to alleviate RAM issues (see if that's too slow though)

[SeguinBe]

This was solved by the replica package, effectively moving to mongo.