Fixed Histogram key_as_string bug

[colings86]

The key as string field in the response for the histogram aggregation will now only show if format is specified on the request.

Closes #6655

[rnonnon-ebiz]

Hi,
Thanks, but that "semi-fix" the problem because Elasticsearch core return the key_as_string, then if you don't use the JAVA API, the problem is still here. Moreover, the JSON returned from Elasticsearch to the JAVA API is still ~1/3 too big.

[s1monw]

@colings86 can we have a test for this too?

[colings86]

@rnonnon I'm not sure I fully understand what you mean. After this change is merged, if you don't specify the format parameter on the histogram aggregation, the `key_as_string' field will not be present in the JSON response and the response will resemble the following:

...
"aggregations": {
      "test": {
         "buckets": [
            {
               "key": 450,
               "doc_count": 23
            },
            {
               "key": 500,
               "doc_count": 2452
            },
            {
               "key": 550,
               "doc_count": 12430
            },
            {
               "key": 600,
               "doc_count": 14982
            },
            {
               "key": 650,
               "doc_count": 13951
            }
            ...

This change does not affect the Java API as the serialisation to the Transpoort layer already does not serialise the key as a string

[colings86]

@s1monw added a rest test to check the JSON output both with and without the format parameter specified

[s1monw]

LGTM

[colings86]

merged but did not auto-close