osd/PeeringState: re-evaluate full OSDs while waiting for recovery reservation

osd/PeeringState: re-evaluate full OSDs while waiting for recovery reservation

A PG can enter *Wait{Local,Remote}RecoveryReserved* before the OSD-map
marks one of its acting OSDs as full, causing us to miss the
recovery_toofull state and stall recovery until a later admin action.

Add react(AdvMap) handlers to both wait states: on every new map,
re-run check_full() and post RecoveryTooFull if needed; otherwise
forward the event unchanged.

Fixes: https://tracker.ceph.com/issues/70670
Signed-off-by: Nitzan Mordechai <nmordec@redhat.com>
(cherry picked from commit b8d2c6832fbd1e630cf820079bc9639c5411772d)