Add support for multi-select facets

[olovy]

Disclaimer: names for these concepts are up for discussion.

Add support for multi-select facets. Normally rendered as check boxes in an interface.
The selected facets from the group are OR-ed.

Make a facet group multi-select by setting connective (as in "logical connective") to OR in _statsrepr

Example

{ "dimensionChain": ["issuanceType"], "itemLimit": 100, "connective": "OR" }

Normally configured in https://github.com/libris/definitions/blob/develop/source/apps.jsonld

In the search response the facet (observation) is marked as selected: true/false and the view-link selects/deselects the facet.
Example: Collection and Integrating are selected.

   "observation": [
      {
        "totalItems": 17299,
        "view": { "@id": "/find?q=*&_limit=20&issuanceType=Collection&issuanceType=Integrating&issuanceType=Monograph" },
        "selected": false,
        "object": { "@id": "https://id.kb.se/vocab/Monograph", ... }
      },
      {
        "totalItems": 10,
        "view": { "@id": "/find?q=*&_limit=20&issuanceType=Integrating" },
        "selected": true,
        "object": { "@id": "https://id.kb.se/vocab/Collection", ... }
      }, 
      ...

The only change needed to the client is to render facets as checkboxes when the selected property is present.

Changes

We've of course always supported filtering on multiple values for the same property. That is in fact the default behavior.
The big changes are in how document counts in facets have to be performed.

ES aggregations are calculated on the query result. We now want to have counts on documents that are outside the result set. For example, when "Serial" is selected we should still show the number of documents with "Monograph". Before this change filters were always applied in the filter section of the query. Multi-select facets/filters have now been moved to post_filter [1] which is run after aggregations. This means we still correct result set in the end.

Since multi-facet filters are applied after aggregation we now have to do some filtering inside the aggregations to get the right count. This is done by moving applying the same filters as in post_filter to each aggregation. Except any filter that refers to the same field.

Elastic is supposedly good at caching filter conditions so this should be efficient.

[1] https://www.elastic.co/guide/en/elasticsearch/reference/current/filter-search-results.html#post-filter

Example

Serial and Monograph are selected

"post_filter": { "bool": { "must": [ { "bool": { "should": [
   { "simple_query_string": { "query": "Serial", "fields": ["issuanceType"], "default_operator": "AND" }},
   { "simple_query_string": { "query": "Monograph", "fields": [ "issuanceType" ], "default_operator": "AND" }
}]}}]}}

same filter on other aggs, for instance meta.bibliography.@id.
a is just a dummy name for the nested agg.

"meta.bibliography.@id": {
  "aggs": {
    "a": {
      "terms": {
        "field": "meta.bibliography.@id",
        "size": 100,
        "order": { "_count": "desc" }
      }
    }
  },
  "filter": { "bool": { "must": [ { "bool": { "should": [
    { "simple_query_string": { "query": "Serial", "fields": ["issuanceType"], "default_operator": "AND" }},
    { "simple_query_string": { "query": "Monograph", "fields": [ "issuanceType" ], "default_operator": "AND" }
  }]}}]}}
}

but not on issuanceType agg

"issuanceType": {
  "aggs": {
    "a": {
      "terms": {
        "field": "issuanceType",
        "size": 100,
        "order": { "_count": "desc" }
      }
    }
  },
  "filter": { "bool": { "must": [] } }
}

Future

This works, but we are reaching the point where the query building has to be refactored. For example not passing around queryParameters everywhere but parse everything into a better representation of the state and requested query.

[lrosenstrom]

Not reviewed in detail but LGTM!

[lrosenstrom]

Also, from a frontend perspective, this is exactly what I need.