semantic_text - document ingestion inference

[carlosdelest]

Adds inference calculation to ingested documents, viaShardBulkInferenceActionFilter.

This action filter intercepts TransportShardBulkActions (before being sent to the data nodes) and calculates inference using the InferenceServiceRegistry / ModelRegistry. Batching is done at the BulkRequest level, so there will be a single inference request for each bulk request (maximum of 512 per request).

There's some nuances with copy_to support, as every source of a copy_to that targets a semantic_text must be inferred as well.

[elasticsearchmachine]

Pinging @elastic/es-search (Team:Search)

[carlosdelest]

Thanks @Mikep86 and @benwtrent for your reviews!

[carlosdelest]

@elasticmachine update branch

[carlosdelest]

Follow-up work to be done in https://github.com/elastic/elasticsearch/pull/108549