feat: better batching in ingest method

[mhordynski]

Feature description

Right now ingest method flow looks like this:

• document processing(chunking) is controlled by ProcessingExecutionStrategy
• All elements are grouped together
• All embeddings are calculated at once

We should also batch requests to embedding model and vector database. One way is to move embedding&inserting into ProcessingExecutionStrategy.

For errors we should decide how to handle failed documents, for example create a default implementation gathering all failed documents into a list. User should be able to create custom error handling strategy.

Motivation

Current behaviour causes problems with rate limits / memory usage / too big requests to embedding models.

Additional context

No response