[Backend] Speed-up ingestion

[galadhremmin]

Eldamo ingestion is still very slow. It can take several hours to ingest the full library, and it's mostly due to the way the repository is implemented and its side-effects.

Some random thoughts:

• Can more be done asynchronously? Currently, indexing and initial ingestion run is done asynchronously, but what else can be moved to asynchronous jobs?
• I'm questioning the use of the Word entity. Can it be denormalized? What's the benefit of keeping it?
• Search indexing and keywords are rather redundant. How can they be optimized to further remove complexity?