We should support optimized targeted arbitrary multi deletes on a sharded timeseries collection by figuring out bucket-level predicate using BucketSpec::createPredicatesOnBucketLevelField helper function so that the shard key extractor can extract proper shard keys for more queries.

For example, predicates on the time field can be translated into predicates on control.min.time since timeseries buckets collection can be sharded based on the time field.

Detailed design section: https://docs.google.com/document/d/1XQHOYrsjxsE9de_xCYyOmZ6AbtgZeXFGDz0r8aQhbbo/edit#heading=h.wk35b7m7wgzl