Closes #840

For example, the result of calling the _cluster/settings API on an off-the-shelf Elasticsearch might look like the following.

GET _cluster/settings?include_defaults | jq '.defaults.cluster.routing.allocation.disk'
{
  "threshold_enabled": "true",
  "watermark": {
    "flood_stage.frozen.max_headroom": "20GB",
    "flood_stage": "95%",
    "high": "90%",
    "low": "85%",
    "enable_for_single_data_node": "false",
    "flood_stage.frozen": "95%"
  },
  "include_relocations": "true",
  "reroute_interval": "60s"
}

When a custom setting related to cluster.routing.allocation.watermark.flood_stage is introduced in transient, the output changes as follows.

GET _cluster/settings?include_defaults | jq '.defaults.cluster.routing.allocation.disk'
{
  "threshold_enabled": "true",
  "watermark": {
    "flood_stage": {
      "frozen": "95%",
      "frozen.max_headroom": "20GB"
    },
    "high": "90%",
    "low": "85%",
    "enable_for_single_data_node": "false"
  },
  "include_relocations": "true",
  "reroute_interval": "60s"
}

As can be seen from this example, the JSON returned by the Elasticsearch _cluster/settings API is not uniform. Therefore, workarounds are necessary to unmarshal it into Go structures. This issue is also pointed out in #509.

This PR addresses the above issue in a general way. When unmarshalling JSON into Go structures, the JSON is converted into a flattened format. The flattened map (i.e., key-value pairs) is then remapped to Go structures using mitchellh/mapstructure. By doing this, it ensures that even non-uniform JSON can be unmarshalled reliably.