[Aggregations] Added _meta construct to inject metadata

[Mpdreamz]

This PR adds the ability to associate a bit of state with each individual aggregation.

Use cases:

• Typed clients can inject some state so that the deserialization
  process has enough information to deserialize the aggregations into
  the correct type. Facets return _type to signal what kind of facet the
  response corresponds with.
• UI state, since aggregations can be nested at arbitrary depths
  injecting pieces of UI state can greatly simplify programming with
  aggregations since there are no two tree data-structures to walk anymore only the aggregation data.

thank you @martijnvg for the help reading the byte[] representation of _meta properly

• this PR also fixes the response format of the new geo_bounds aggregation to wrap the result in a "aggregationname" : {}

Here's a gist of a sense session you can use to review this branch

Example request:

{
"aggs": {
    "name": {
      "terms": {
        "field": "title"
      },
      "meta": 2,
      "aggs": {
        "viewport" : {
          "meta" : {
               "complex" : "object"
          },
          "geo_bounds" : {
            "field" : "location"
          }
        },
        "empty_agg" : {
          "meta" : "should still be returned",
          "sum": {
            "field": "i_do_not_exist"
          }
        },

Example response:

{
   "aggregations": {
      "name": {
         "meta": 2,
         "buckets": [
            {
               "key": "banner",
               "doc_count": 1,
               "viewport": {
                  "meta": {
                    "complex": "object"
                  },
                  "bounds": {
                     "top_left": {
                        "lat": 28,
                        "lon": 21
                     },
                     "bottom_right": {
                        "lat": 28,
                        "lon": 21
                     }
                  }
               },
               "nested_terms": {
                  "meta": {
                     "complex": "object",
                     "nuber": 1
                  },
                  "buckets": [
                     {
                        "key": "a",
                        "doc_count": 1,
                        "counter": {
                           "meta": 4.1,
                           "value": 40
                        }
                     },

[uboness]

let's get rid of the underscore... we moved away from that (no real need for it as all the keys under the agg are controlled by us)

[Mpdreamz]

Thanks for review @uboness ! I wrapped the create() because I wasn't sure if updating all the Aggregation implementation constructors would break elasticsearch's public API.

I opted for _meta because adding it in the response might break existing code that uses meta as aggregation name.

Will update both property names to meta and pass metaData in the constructors of Aggregation implementations.

[uboness]

why would it break? it's under the agg name, isn't it? in any case... still... not good enough reason to introduce inconsistency in the api. That is, if we feel that something will break apis, better break them and target for a later release (one that enables breaking changes) rather than introduce inconsistency in the api (something we'll need to cary later for a long time)

[Mpdreamz]

@uboness I've updated the PR with all your feedback I only have one question regarding the AggregationBuilder.java changes.

[jpountz]

the way that we usually do this is by making toXContent final, write the name and metadata, and delegate to a doXContentBody method. This ensures that no aggregation can by-pass the serialization of the metadata.

[jpountz]

cc @colings86

[clintongormley]

@Mpdreamz where are you on this one?

[Mpdreamz]

closing in favour of #8279