[ML] Fix issues in dynamically reading the number of allocations

[davidkyle]

Relates to problems in the GET inference API which should dynamically update the num_allocations field with the actual number from the deployed model. This is required for adaptive allocations where the field will change dynamically.

The num_allocations field in service_settings is updated with the current value.

GET _inference/elser_on_ml

{
  "endpoints": [
    {
      "inference_id": "elser_on_ml",
      "task_type": "sparse_embedding",
      "service": "elasticsearch",
      "service_settings": {
        "num_allocations": 3,            <-- this field is dynamic
        "num_threads": 1,
        "model_id": ".elser_model_2",
        "deployment_id": ".elser_model_2_for_me"
      },
      "chunking_settings": {
        "strategy": "sentence",
        "max_chunk_size": 250,
        "sentence_overlap": 1
      }
    }
  ]
}

The first issue is that GroupedActionListener throws if called with size == 0. This is now protected against by skipping the model update if the list is empty.

The second issue is that the wrong field was being updated so the update was not seen in the API response. Tests are added to cover both cases.

Non issue as the code is not live

[elasticsearchmachine]

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

[davidkyle]

Without this check the GroupedActionListner was called with 0 requests which throws an exception