Avoid using the same scheduler for compaction and flush

[evenyag]

What type of enhancement is this?

Refactor, Tech debt reduction

What does the enhancement do?

Although compaction executes in a different runtime, it share the same scheduler with flush jobs. So it may blocks flush jobs as they share the same max_background_jobs limit. #4804 can't fully solved this.

A better approach may be separating schedulers for compaction and flush.

Implementation challenges

We need to add a new scheduler for flush and use different limit to control the concurrency of compaction and flush. We might also refactor how compaction spawns tasks.