Improve long-term scalability of CancelRedundantPipelinesService

Summary

In #414833 (closed), we made found a better index to query and shortened the cancelable pipeline window from 7 to 3 days, due to the query performance being bad enough to actually degrade GitLab SaaS.

In addition to the relatively narrow fixes, there was discussion of longer-term scalability solutions: #414833 (comment 1485475470)

Those should be further investigated so we can continue to provide this feature with as-good or better performance as we scale.

Improvements

• Decrease the duration of database queries from CancelRedundantPipelineService

Risks

• We could accidentally make performance worse
• We could accidentally make the feature less reliable

Involved components

• CancelRedundantPipelineService

Proposals

• #414833 (comment 1485475470)
• Use the current query, but execute it multiple times for different time windows (timestamp-based batching strategy)