[ML] adds geo_centroid aggregation support to data frames

[benwtrent]

This adds support for the geo_centroid aggregation. The mapped type that supports geo_centroid is always geo_point. So, I opted to make that the static destination mapped type as well. ES supports geo_point _sources being index in the form of lat, lon so calling .toString on the aggregation results is good enough to build the document source.

This opens up some interesting use cases to show the centroid for specific entities when the data is pivoted. An example using kibana_sample_data_logs

PUT _data_frame/transforms/geo_test
{
  "source": {
    "index": "kibana_sample_data_logs"
  }
  ,"dest": {"index": "geo_test"}
  , "pivot": {
    "group_by": {
      "clientip": {"terms": {"field": "clientip"}}
    },
    "aggregations": {
      "location": {
        "geo_centroid": {
          "field":"geo.coordinates"
        }
      }
    }
  }
}

[elasticmachine]

Pinging @elastic/ml-core

[droberts195]

LGTM