Describe the bug

The rr monitor TUI displays a "Latency" metric that conflates actual network/SSH latency with the time required to collect metrics on the remote host. On
macOS, network bandwidth calculation uses nettop -L N which samples over N seconds, causing the reported "latency" to show ~5 seconds even when actual
SSH latency is under 100ms.

This is misleading because users see "5173ms degraded" and assume there's a network problem when the connection is actually healthy.

To reproduce

1. Configure two macOS hosts in ~/.rr/config.yaml
2. Run rr monitor
3. Observe "Latency" showing ~5000ms+ with "degraded" status
4. Verify actual latency is fine:

   # SSH roundtrip is ~100ms                                                                                                                               
   time ssh <host> "echo ok"                                                                                                                               
                                                                                                                                                           
   # rr status shows correct latency                                                                                                                       
   rr status --machine                                                                                                                                     
   # Returns: "latency": "75.497083ms"                                                                                                                     
                                                                                                                                                           

Expected behavior

The "Latency" metric should reflect actual network/SSH latency (what rr status reports), not the total metrics collection time. If metrics collection takes
longer due to sampling windows, that duration should either:

1. Not be included in the latency display, or
2. Be shown as a separate "Collection Time" metric, or
3. Be documented/labeled clearly (e.g., "Probe Time" instead of "Latency")

The "degraded" threshold should apply to actual network latency, not collection time.

Environment

• OS: macOS 15.7.3 (also affects remote macOS hosts)
• rr version: v0.14.1 (commit b2b1ac7)
• Go version: go1.24.11
• Architecture: darwin/arm64

Additional context

The root cause appears to be nettop usage for network bandwidth stats. To calculate bytes/sec, multiple samples are needed:

  # Single sample - fast (~150ms)                                                                                                                            
  time ssh host 'nettop -P -L 1 -J bytes_in,bytes_out'                                                                                                     
                                                                                                                                                             
  # 5 samples - slow (~4-5s)                                                                                                                                 
  time ssh host 'nettop -P -L 5 -J bytes_in,bytes_out'                                                         

rr status --machine correctly reports SSH probe latency (~75-100ms), so the accurate measurement exists. It's just not what rr monitor displays.

Relevant config:

  # ~/.rr/config.yaml                                                                                                                                        
  hosts:                                                                                                                                                     
    congress-m2-ultra:                                                                                                                                       
      ssh:                                                                                                                                                   
        - mac-m4                                                                                                                                           
      dir: ~/rr/${PROJECT}                                                                                                                                   
    lavaka-m2-ultra:                                                                                                                                         
      ssh:                                                                                                                                                   
        - mac-m1                                                                                                                                             
      dir: ~/rr/${PROJECT}                                                                                                                                   

Both hosts show identical ~5170ms "latency" in monitor, but rr status shows 75ms and 100ms respectively.