[FEATURE] Document ingestion with offline batch inference

[heemin32]

Is your feature request related to a problem?
Following up on #2891, I’d like to request support for the asynchronous batch ingestion API to handle embedding ingestion after inference is completed, instead of requiring the client to perform the ingestion.

What solution would you like?
An API that automatically ingests embeddings once they are generated by the model.

What alternatives have you considered?
Allowing the client to handle the embedding ingestion after the embeddings are generated.

Do you have any additional context?
N/A