[feat] Better approach for rate-limiting embedding queries

[triathematician]

Current method:

• batch process embedding queries by document
• could have 100s+ chunks in a single embedding API call

Proposed method:

• give embedding models a batch limit for processing
• (potentially) enforce a delay between queries

[triathematician]

Will enforce this with TextDocEmbeddings rather than in the embedding service API. Clients of an EmbeddingService will be responsible for managing API usage appropriately.