Expertise API/Image: Cache or create a store of embeddings

[haroldrubio]

We previously cached embeddings in an addon module for Redis but ran into some issues with it failing to fetch.

Now that we are executing the model serverless, we need a way for an expertise container to pull embeddings from, most likely Google Cloud, into its local filesystem.

A model-agnostic implementation may be difficult since the embedding step happens at the model level. This might have to happen in the create_dataset step and only embed publications in the filesystem, and merge the 2 embedding files.