Drift between `total_queries` and `finished_queries` metrics in status when setting watches from multiple clients

If you run two parallel instances of this test program against a fresh fdb database and then, any amount of time later, check its status.json, you'll see something like this snippet:

"cluster": {
  "processes": {
    "xyz": {
      "roles": [
        {
          "role": "storage",
          "total_queries": {
            "counter": 3082050,
            "hz": 15.1999,
            "roughness": 6.7680300000000004
          },
          "finished_queries": {
            "counter": 3082015,
            "hz": 15.1999,
            "roughness": 6.7680300000000004
          },
          "query_queue_max": 41,
        }
      ]
    }
  }
}

This difference between total_queries and finished_queries will remain elevated for the life of the storage process (and will get larger every time you run two instances of the test program). query_queue_max, since it is computed based on the difference between those two counters, will likewise go up and never come back down.

This is fixed in #13786 .