Scrub reserves from actingbackfill put waits for acting
Scrub sends requests to actingbackfill shards, but waits for only acting.size() grants. This happens when a PG is remapped causing scrub to start possibly before granted a reservation from all acting which is what is required. This causes an extra PG::sched_scrub() call which finds the PG already scrubbing.
I added ceph_assert(!is_scrubbing()); to PG::sched_scrub() which triggered multiple times in a rados suite run.