save base embeddings to disk, recalculate periodically?

[julianfleck]

With the knowledge graph growing, starting up the API becomes slow because the computation of the base embeddings takes a lot of time. Instead, we should save then somewhere and load then from disk instead of recalculating everytime. Then we can run the calculations in the background (ideally with a scope parameter, e.g. ECCHR as a scope limits the computation of the embeddings to a certain part of the map).