Store vector embeddings

[mihow]

Summary

Add support for saving vector embeddings from model output to be used for unknown species detection & similarity search.

Implementation Details

We can begin by returning and storing vectors for each classification for each detection in the same way we are returning & storing logits in the postgres database and querying with pgvector. But consider exploring adding a new dedicated vector database to the stack.

Need support for multiple embeddings! but is one per each classification/model output okay? When we get to querying/using the vectors, that is the point when we need select which embedding (or the model that made them if its one-to-one)

Related Issues

...

Additional Context

...

[mihow]

Approaches to test:

• Add pgvector extension to our Antenna postgres database & add similarity search queries to the Django ORM

[mihow]

Closed by #821