[Feature Request] Allow custom min_doc_count in queries

[hank-sq]

Summary

We often make queries like 'give me all sales for this specific month grouped by day'. When we do this, it would be helpful to still return a node even when there are no documents. Providing these empty nodes is still helpful for our clients, who maybe want to still display a point in a graph with a 0 value.

It looks like the min_doc_count is hard coded as 1 to intentionally not return empty buckets, but having the ability to override this and return empty buckets would be very helpful!

[myronmarston]

I saw you got the ball rolling on this in #643--thanks for that! While you're requesting this feature for use on date histogram groupings (so you can lower min_doc_count to 0), I think this would be useful on every groupedBy field, and it's also available on term groupings. As an example, I've sometimes wanted to identify which groupable id has the most underlying documents in its bucket. We don't support a way to sort aggregations on document counts so there's no easy way to determine that today. This option would provide a way to determine that using adhoc querying--you could repeatedly run an aggregation query with different large values of min_doc_count to identify the largest buckets.

The EG Query API is intentionally designed to not be a "verbatim" translation of the underlying Elasticsearch API. The fact that an ElasticGraph application uses OpenSearch/Elasticsearch under the covers is an implementation detail, and not something that we want the GraphQL schema to necessarily advertise.

So, we should consider what we want to call this option. If we think min_doc_count is the ideal name for it, let's stick with it, but we shouldn't choose it as the name just because that's what Elasticsearch chose.

We also need to consider if we want to set shard_min_doc_count as it's related.

[hank-sq]

@myronmarston my specific need is for groupings on date times, to be able to include empty periods and also set the 'extend_bounds' to include empty groupings throughout the whole period I'm interested in.

So specifically, I don't have a need to override the minimum count other than being able to request empty buckets.

I'm not sure if you want to consider that the same problem or separate (i.e. we could expose a 'include_empty' parameter that is a boolean and not the raw min_doc_count specifically for date_histograms and solve the generalized term_grouping minimum node separately).

Alternatively, if there's a generalized solution that would empower getting empty buckets that's fine too.