Prefilter GROUPBY LIMIT queries

[kaikalur]

We have seen users running ad-hoc queries like:

SELECT SUM(x), userid FROM Table GROUP BY userid LIMIT 1000

if the Table has a large number of distinct userid, these queries often OOM or timeout. So we added a new optimization that first tries to scan the table looking for LIMIT number of distinct keys (in a short time like 10s, configurable) and if it finds them, makes a map of those and filters out the original table using that. Roughly:

SELECT SUM(x) , userid FROM Table
CROSS JOIN (SELECT MAP_AGG(hash(userid)) m FROM (SELECT DISTINCT userid FROM Table LIMIT 1000)))
WHERE IF(CARDINALITY(m)=1000, m[hash(userid)], TRUE)

So the idea is if there are 1000 keys and if we find them quickly then those are returned for quick lookup (prefilter), if not there are no 1000 distinct keys or if we could not find them within the timeout limit, this filter becomes a no-op.

Test plan - AbstractTestQueries

== RELEASE NOTES ==

General Changes
* Added a new optimization for filtering large tables with LIMIT number of keys for queries that do simple GROUP BY LIMIT with no ORDER BY. Added a boolean session param: `prefilter_for_groupby_limit` that can enable this feature

[jainxrohit]

Can we change commit message to imperative style?

[kaikalur]

Can we change commit message to imperative style?

Done