osd/PeeringState: drop pg_rebuild_max_secs/min_secs from the interim rebuild-stats solution

pg_rebuild_duration's avgtime (sum/avgcount) is sufficient on its own
for the durability-score use case these counters were added for; the
max/min-seconds counters are just supplementary statistics, not required
for that calculation.

Since recoverystate_perf is a single OSD-wide PerfCounters object shared
by every PG shard, and the max/min update logic used a caller-side
get()+compare+set() sequence that is not atomic as a whole (unlike
tinc()'s true atomic accumulation for sum/avgcount), so concurrently
recording PGs could race and silently lose an update to these min/max
counters. These can be fixed, but is deferred until the full solution.

Remove rs_pg_rebuild_max_secs/rs_pg_rebuild_min_secs (enum entries and
registration), the counter updates in prepare_stats_for_publish(), and
the corresponding assertions in TestPeeringState.cc. Only
rs_pg_rebuild_duration remains.

Fixes: https://tracker.ceph.com/issues/79028
Signed-off-by: Sridhar Seshasayee <sridhar.seshasayee@ibm.com>