Index-time inference in ElasticsearchDocumentStore (optional / deferred)

[davidsbatista]

• Add optional support for using an Elasticsearch ingest pipeline that runs an inference processor (e.g. ELSER or dense model) so document embeddings are produced in ES at index time.
• New optional init parameter(s), e.g. inference_ingest_pipeline_id or similar; default behaviour unchanged.
• ES: Inference ingest processor, Ingest pipelines.