[FEATURE] Show only intersecting buckets to the Adjacency matrix aggregation

[brusic]

Is your feature request related to a problem? Please describe.

The Adjacency matrix aggregation will compute and return all buckets related to any of the combinations of the filters specified, including hits on a single filter. However, sometimes a user would want to return only the buckets related to an intersection and not for a single filter. An optional show_only_intersecting parameter would stop buckets resulting from a single filter being hit to be created.

The original intent of this feature was to support a minimum_should_match analogous to boolean queries, but the aggregation only returns adjacencies and not the combination of every filter, so the result could be confusing.

Describe the solution you'd like

From the official documentation, with the optional minimum_should_match parameter added:

GET opensearch_dashboards_sample_data_ecommerce/_search
{
  "size": 0,
  "aggs": {
    "interactions": {
      "adjacency_matrix": {
        "show_only_intersecting": true,
        "filters": {
          "grpA": {
            "match": {
              "manufacturer.keyword": "Low Tide Media"
            }
          },
          "grpB": {
            "match": {
              "manufacturer.keyword": "Elitelligence"
            }
          },
          "grpC": {
            "match": {
              "manufacturer.keyword": "Oceanavigations"
            }
          }
        }
      }
    }
  }
}

New response:

{
   ...
   "aggregations" : {
     "interactions" : {
       "buckets" : [
         {
           "key" : "grpA&grpB",
           "doc_count" : 590
         },
         {
           "key" : "grpA&grpC",
           "doc_count" : 329
         },
         {
           "key" : "grpB&grpC",
           "doc_count" : 299
         }
       ]
     }
   }
 }

Buckets grpA, grpB, and grpC have been omitted.

Describe alternatives you've considered

The existing solutions would be to return all buckets and then use a pipeline aggregation to reduce the number of buckets, or prune in application code. Both solutions work post-reduce, after the creation of buckets, so there is needless overhead involved. The coordinating node could see too many bucket errors, especially when used in conjunction with sub-aggregations.

Additional context

I already have a PR with a working solution ready to be reviewed if this feature is deemed to be useful. The original implementation to support minimum_should_match was more detailed, but now it would be a single if statement.