Faster filtered scroll

[generall]

All Submissions:

• old scroll implementation did a full scan, if the filter had low cardinality, which resulted in very high latency of the scroll query
• proposed implementation works similar to query planner, it estimates the cardinality of the filter first and then decides if it makes sense to use index first or filter the ids.
• This implementation resolves the case, where user wants to select some records from DB using the secondary index.

Our of scope:

current boundary threshold is hard-coded and naive. It would require additional telemetry data to decide how to adjust it.

Issue detected in demo by @joein