[ML] add new trained_models/{model_id}/_infer endpoint for all supervised models and deprecate deployment infer api

[benwtrent]

This commit adds a new _ml/trained_models/{model_id}/_infer API. This api works for both native NLP models and supervised models trained via Data Frame analytics.

The format of the API is the same as the old _ml/trained_models/{model_id}/deployment/_infer. Taking a docs and an inference_config parameter.

This PR also deprecates the old experimental _ml/trained_models/{model_id}/deployment/_infer API.

The biggest difference is that the response now nests all results under an "inference_results" object.

closes: #86032

[elasticmachine]

Pinging @elastic/clients-team (Team:Clients)

[elasticmachine]

Pinging @elastic/ml-core (Team:ML)

[elasticsearchmachine]

Hi @benwtrent, I've created a changelog YAML for you.

[benwtrent]

Should wait until all the work with: elastic/ml-cpp#2258 is completed and pytorch tests unmuted. That way we can fully test functionality in upgrade tests and integration tests.

[benwtrent]

We needed an external name for interaction at the REST layer and since inference needs to work during rolling upgrade, we cannot get rid of the old action name.

So, both point to effectively the same transport class.

[sethmlarson]

LGTM from an API perspective

[benwtrent]

@elasticmachine update branch

[davidkyle]

LGTM

[davidkyle]

Can you add an example using the built in lang_ident model pls

[davidkyle]

Alternatively inferenceResults which implement ToXContentFragment could be mapped to ToXContentObject

        static ToXContentObject toXContentObject(ToXContentFragment fragment) {
            return (builder, params) -> {
                builder.startObject();
                fragment.toXContent(builder, params);
                builder.endObject();
                return builder;
            };
        }

and

builder.field("inference_results", inferenceResults.stream().map(Response::toXContentObject).collect(Collectors.toList()));

[lcawl]

If we're keeping both the old and new doc pages (with the old API marked as deprecated), shouldn't the same be done for the rest-api-spec/src/main/resources/rest-api-spec/api/ml.infer_trained_model.json? (i.e. leave the existing .json file and just add the "deprecated" property). That way when we get to the point of generating docs from those specs, it'll match.

[benwtrent]

That way when we get to the point of generating docs from those specs, it'll match.

I don't know. We don't have a deprecated flag to allow multiple endpoints in the same doc, but we do have that for the spec. Regardless, these are still experimental APIs.