feat: EmbeddingsBuilder redesign

[garance-buricatu]

Feature Request

Any object that implements the trait Embedabble can be added to the embeddings builder and therefore can be embedded.

Motivation

Currently, embeddings builder is tightly coupled with the DocumentEmbedding type (only objects of this type can be used by the builder). Downsides: user needs to type cast their types to DocumentEmbedding and the structure of DocumentEmbedding is predetermined so the user needs to design their embedding code around this.

Proposal

1. Define the trait Embedabble which contains a method fn embedabble() which returns a list of strings (the strings that should be embedded).
2. Change the EmbeddingsBuilder to accept any type that implements the trait Embedabble.
3. Write a custom derive macro that can be used on any struct which derives the Embedabble trait automatically.

Linked issues:

#60
#61
#62

Todo:

• Give trait and macro same name like serde

• implement Embeddable trait for many primitive types

• macro fails when there are no embed tags

• macro fails when embed with is empty string

• need to handle error in custom embed function

• review borrow and owning eveywhere. done for embeddings.rs

• update docstrings. done for embeddings.rs

• clean up embedding builder definition with traits maybe?

• should build() return Embedding object or just string? return Embedding object - document string used for debugging

• in memory vector store, rethink type of document D --> move to another branch

• why is vector search of in memory vector store so bad? bug found

• in memory store - implement from trait from both returns types of build into add_document

• tests

• make macro as feature flag

[cvauclair]

@garance-buricatu can you add some formatting and a little bit more context to this issue? Thank you!