refactor: Update VectorStoreIndex trait

[marieaurore123]

Github issue: #41

• Refactor VectorStoreIndex trait

Notes:

• kept DocumentEmbeddings as the InMemoryVectorStore value type. Did this because if I change it to a more general type, I would need to change the VectorStore trait which should be done in a separate branch.
• method top_n_ids of VectorStoreIndex contains a generic type: <T: for<'a> Deserialize<'a>. Technically, it we don't need type T to get the ids of the top n documents. However, I found it simpler to get the ids of the top n by calling the existing method top_n which needs to be called like this top_n::<T>() and filtering the ids only. This also allows the implementation of the method top_n_ids to be to same for all trait implementations (ie. implementation defined in trait def)

Please let me know what you think about the second point! Thanks:)

[0xMochan]

Seems like a lot of code got simplified 👍

[cvauclair]

@garance-buricatu for the same of record keeping, here are my opinions on your points:

kept DocumentEmbeddings as the InMemoryVectorStore value type. Did this because if I change it to a more general type, I would need to change the VectorStore trait which should be done in a separate branch.

That's fine for now, we'll remove the DocumentEmbeddings type in a subsequent PR (or maybe keep it only for the internal representation of the InMemoryVectorStore)

method top_n_ids of VectorStoreIndex contains a generic type: <T: for<'a> Deserialize<'a>. Technically, it we don't need type T to get the ids of the top n documents. However, I found it simpler to get the ids of the top n by calling the existing method top_n which needs to be called like this top_n::() and filtering the ids only. This also allows the implementation of the method top_n_ids to be to same for all trait implementations (ie. implementation defined in trait def)

IMO we should remove the provided implementation since 1) some DBs might have a more optimized way to query the ids only (e.g.: SELECT id FROM ... instead of SELECT * FROM ...); and 2) we don't know the name of the primary key used by the documents in advance.