max-pg-per-osd.from-primary fails because of activating pg
the reason why we have activating pg when the number of pg is under the hard limit of max-pg-per-osd is that:
1. osd.1 received the osdmap instructing it to create pg 1.0, so it updated its replica osd.2 to create pg 1.0
2. osd.2 is capped by the max-pg-per-osd when it's about to create pg 1.0. so it dropped the request to create the pg.
3. the pool 15 gets removed in osdmap#58.
4. but the updated osdmap is not sent to osd.1 or osd.2 before the wait_for_clean() times out.