PGMap::recovery_rate_summary() summaries the recovery progress from
a pool's
pool_stat_t.stats.sum.num_{objects_recovered,bytes_recovered,keys_recovered},
now we only increment stats on completion of recovery push, but there
are chances that a PG recovers by pulling data from replicas to primary.
in that case, teuthology will erroneously consider recovery hung: a zero
recovering_keys_per_sec or recovering_bytes_per_sec or
recovering_objects_per_sec. so we should increment stats on pull as
well.