fix(replication): bin transfer summaries at 128 MiB and keep window refresh off the hot path

Second review round:

- update_xfer_rate split at 1 MiB while the minio-go transferSummary
  labels (and RustFS's own worker-pool split) mean >= 128 MiB for
  Large, so a 2 MiB replication reported under Large with Small stuck
  at zero. The producer now bins on MIN_LARGE_OBJ_SIZE; a MetricsV2
  assertion covers 2 MiB / 127 MiB / exactly 128 MiB.
- add_size no longer recomputes the rolling windows: two full
  one-hour-deque scans per failure under the bucket-stats write lock
  made failure bursts quadratic (30k events ~2.1s). The windows are
  stamped only at the collection point (get_latest_replication_stats,
  which serves both the local leg and the peer RPC); the aggregation
  regression now drives that path explicitly before the RPC round trip
  and merge.

crates/replication/src/stats.rs

@@ -543,13 +543,14 @@ impl FailStats {
         self.size = self.size.saturating_add(size);
         self.recent.push_back(FailureSample { observed_at, size });
         self.prune(observed_at);
-        self.refresh_windows();
     }
 
     /// Recompute the serializable rolling-window snapshots from the local
-    /// samples. Only meaningful on the live per-node struct: a deserialized
-    /// or merged struct has no samples, and refreshing it would wipe the
-    /// aggregated windows.
+    /// samples. Called at the collection point (per-node stats snapshot),
+    /// never on the failure hot path — the two deque scans are O(window) and
+    /// `add_size` runs under the bucket-stats write lock. Only meaningful on
+    /// the live per-node struct: a deserialized or merged struct has no
+    /// samples, and refreshing it would wipe the aggregated windows.
     pub fn refresh_windows(&mut self) {
         self.last_minute = self.recent_since(Duration::from_secs(60));
         self.last_hour = self.recent_since(Duration::from_secs(3600));
@@ -664,7 +665,9 @@ impl BucketReplicationStat {
     }
 
     pub fn update_xfer_rate(&mut self, size: i64, duration: Duration) {
-        if size > 1024 * 1024 {
+        // Same boundary as the worker-pool split and minio-go's
+        // Large/Small transfer-summary labels: >= 128 MiB is "large".
+        if size >= crate::runtime::MIN_LARGE_OBJ_SIZE {
             self.xfer_rate_lrg.add_size(size, duration);
         } else {
             self.xfer_rate_sml.add_size(size, duration);