SQL query uses TopN when grouping by time and other dimension

### Description

When querying the SQL druid endpoint with a query doing a group-by on two fields and on of them is time related, the TopN query-type could be used instead of the group-by one, with the time grouping being implemented as "granularity".

Example:
```
EXPLAIN PLAN FOR
SELECT FLOOR("__time" TO MONTH) AS "__timestamp",
"my_field" AS "my_field",
SUM(my_value) AS "sum_my_value"
FROM my_data_source
WHERE "__time" >= '...'
AND "__time" < '...'
GROUP BY "my_field", "FLOOR("__time" TO MONTH)
ORDER BY sum_my_value
LIMIT 500;

DruidQueryRel(query=[{"queryType":"groupBy","dataSource":{"type":"table","name":"my_data_source"}...
```

### Motivation

Group-by queries are a lot more expensive than TopN queries, this change would allow to get results a lot faster and cheaper.

Start with the SQL Druid endpoint's query-planning path and the EXPLAIN PLAN example in the issue, focusing on how group-by queries select their query type and represent time grouping. Done means eligible groupings use a TopN query with time represented as granularity, while preserving the requested grouped results and ordering.

Written by the indexing model from the issue text.