Aggregations: Allow calculation of bucket fields

[max-moser]

In InvenioRDM, we would like to distinguish between record-view events that go through the API or UI via a boolean field via_api in the event.
Currently, during aggregation we can either discard events with either of the values (via_api=true/false), or discard the information from this field altogether – or we can cobble together a unique_id that separates the two kinds of events (e.g. {prefix}-{recid} where prefix would be either "ui" or "api").
In the statistics PR in app-rdm (inveniosoftware/invenio-app-rdm#2080), we went the third route for now.

Neither of these possibilities is too enticing, we would rather be able to aggregate the document counts per value inside the aggregation, similar to:

{
  "via_api": [
    {"key": true, "doc_count": 123},
    {"key": false, "doc_count": 567},
  ]
}

This would probably require a bucket_fields feature in the aggregations, similar to the metric_fields.