Create a multi-DB wrapper for giant datasets

[sdenton4]

Truly gigantic datasets like the A2O get too large for a single vector database.

First, insertion of new vectors into usearch DB slows down steadily as the DB increases in size. Second, the gigantic databases themselves become unwieldy to move around, and SQLite has a hard time with cloud resources. Third, large datasets generally require parallelization for ingestion, which implies the creation of multiple databases which must eventually be merged or else wrapped.

[sdenton4]

Suggested strategy:

• Make a db/multi_db_impl.py which implements the HopliteDBInterface, and wraps multiple HopliteDBInterface instances.
• It's ok if the MultiDB is read-only, though it's worth thinking about how we might lift that restriction (especially for adding more annotations/labels).
• The HopliteDBInterface relies on int ID numbers for many operations. These are intentionally 'non-informative' and there should be no user expectations that particular ID numbers remain constant across different implementations, etc. We will need some logic in the MultiDB wrapper for translating ID numbers into some 'common' space as the raw ints likely collide across the multiple wrapped databases.