Same memory when geo aggregations are not on top

[nik9000]

Saves memory when the geotile_grid and geohash_grid are not on the
top level by using the LongKeyedBucketOrds we built in #55873.

[elasticmachine]

Pinging @elastic/es-analytics-geo (:Analytics/Aggregations)

[nik9000]

I'd love to build a performance test for this one too! My desktop is busy as the moment though.

[nik9000]

Rally before and after.

Highlights:

| before | 90th percentile service time | date-histo-geohash-grid |     4595.70 |     ms |
|  after | 90th percentile service time | date-histo-geohash-grid |     2836.80 |     ms |
| before | 90th percentile service time | date-histo-geotile-grid |     4831.04 |     ms |
|  after | 90th percentile service time | date-histo-geotile-grid |     2885.81 |     ms |

That is about 40% performance improvement in my test case. It looks like:

curl -XPOST -HContent-Type:application/json localhost:9200/noaa/_search -d'{
  "size": 0,
  "query": {
    "geo_bounding_box": {
      "station.location": {
        "top_left": "c",
        "bottom_right": "9"
      }
    }
  },
  "aggs": {
    "date": {
      "date_histogram": {
        "field": "date",
        "calendar_interval": "1w"
      },
      "aggs": {
        "location": {
          "geotile_grid": {
            "field": "station.location",
            "precision": 2
          }
        }
      }
    }
  }
}'

[talevy]

wow. look at that benchmark! 😮 🎈