[ML] Associate text_expansion subsearch with parent search request

[prwhelan]

Description

A text_expansion subsearch is hardcoded to time out after 10s, ignoring any timeout configured in the parent search task.

For example:

GET kibana_sample_data_logs/_search
{
  "timeout": "30s",
  "query": {
    "text_expansion": {
      "message": {
        "model_id": ".elser_model_2",
        "model_text": "the query string"
      }
    }
  }
}

If the text_expansion takes longer than 10s, it will time out and throw an error, ignoring the 30s configuration.

Options:

1. Pass the configured timeout value from the parent search to the subsearches: prwhelan@241296d
2. Associate the subsearch task with the parent task and remove the hardcoded timeout from the test_expansion task. When the parent task is canceled, all children will be canceled.

[elasticsearchmachine]

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

[Mikep86]

Update on this after talking with @jimczi:

The timeout parameter at the root level of the search request is only intended to control timeouts getting results from shards, potentially allowing for partial search results if desired. Overloading this parameter to control inference timeouts is not appropriate.

Instead, a proposed solution to configure query-time inference timeouts is to add an index setting for this.

[davidkyle]

Closing as #131551 covers the same problem