Prefer backfill for inactive PGs over recovery
If number of active replicas in PG drops below min_size, request to that PG are blocked and PG waits for backfill.
If there's a lot of log-based recovery at the same time happening in the cluster, PGs may be blocked for a significant amount of time.
It would be beneficial to prefer backfill of inactive PGs so that the cluster can quicker become operational.
Note: This ticket has already been implemented in: https://github.com/ceph/ceph/pull/12389 and has been added to enable backport.