[9.2] Add new parameters to inference PUT API for the rerank task type

[kosabogi]

Related dev issue: elastic/elasticsearch#130485

Starting in version 9.2, the PUT _inference/{task_type}/{elasticsearch_inference_id}
API with the rerank task type introduces two new parameters in service_settings:

• long_document_strategy
• max_chunks_per_doc

These parameters define how the reranker handles long documents during inference.

long_document_strategy

Type: string (optional)

Controls the strategy used for processing long documents.

Possible values:

• truncate (default): Processes only the beginning of each document.
• chunk: Splits long documents into smaller parts (chunks) before inference.

To enable chunking, this value must be set to "chunk".

max_chunks_per_doc

Type: integer (optional)

Limits how many chunks per document are sent for inference when chunking is enabled.
If not set, all chunks generated for the document are processed.

Required updates

• Add long_document_strategy and max_chunks_per_doc to the service_settings object of the PUT inference API documentation

• Update the Configuring chunking documentation to mention that chunking must be enabled by setting:

"long_document_strategy": "chunk"

• Update the description of the chunking_settings object to clarify that enabling chunking requires setting the long_document_strategy parameter to chunk in the rerank inference endpoint configuration.