Accept LTR inference config when creating model

[demjened]

This PR adds support for importing a model with a specific inference_config. This allows us to use regression models with the learning_to_rank inference type that come with externally supplied configuration.

Example:

MLModel.import_model(
    es_client,
    model_id,
    regressor,
    feature_names,
    es_if_exists="replace",
    es_compress_model_definition=compress_model_definition,
    inference_config={
        "learning_to_rank": {
            "feature_extractors": [
                {
                    "query_extractor": {
                        "feature_name": "title_bm25",
                        "query": {"match": {"title": "{{query_string}}"}},
                    }
                },
                {
                    "query_extractor": {
                        "feature_name": "imdb_rating",
                        "query": {
                            "script_score": {
                                "query": {"exists": {"field": "imdbRating"}},
                                "script": {"source": 'return doc["imdbRating"].value;'},
                            }
                        },
                    }
                },
            ]
        }
    },
)

The change is backward compatible - if inference_config is not passed, it will be set to the default {<model_type>: {}}.