[POC] Support text similarity reranking based on best-scoring snippet

[kderusso]

Example:

POST my-index/_search?error_trace
{
  "track_total_hits": true,
  "fields": [
    "text"
  ],
  "retriever": {
    "text_similarity_reranker": {
      "retriever": {
        "standard": {
          "query": {
            "match": {
              "text": "What is the capital of the USA?"
            }
          }
        }
      },
      "field": "text",
      "snippets": {
        "num_fragments": 3,
        "max_size": 20
      },
      "rank_window_size": 10,
      "inference_id": "my-rerank-model",
      "inference_text": "What is the capital of the USA?"
    }
  }
}

[kderusso]

Closing this POC PR. It was useful for discussion, but the final solution will not involve modifying the retriever API to specify snippet information in this manner