feat(T3.3): cpu load readout + lifeguard health score

rpc.rs: Add read_cpu_load_1m() (/proc/loadavg), compute_health_score()
(hot>90% +1, fs>90% +1, load>4 +1, failed timers +1), expose both in
DashboardStatusReply. NodeStatusV2 and api.ts propagated.

NodeCard: cpu load row in facts grid (amber when >4), healthOf() now
driven by server-side health_score (>= 3 → err, 1-2 → warn) with
client-side fs/timer fallback for older daemons.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

claw-store/src/serve_v2.rs

@@ -368,6 +368,12 @@ pub struct NodeStatusV2 {
     /// Unix timestamp (seconds) when the peer's daemon last started.
     #[serde(default)]
     pub daemon_started_unix: Option<u64>,
+    /// 1-minute load average. None on non-Linux or when the peer is offline.
+    #[serde(default)]
+    pub cpu_load_1m: Option<f32>,
+    /// Lifeguard-style health score: 0 = healthy, higher = stressed (T3.3).
+    #[serde(default)]
+    pub health_score: u8,
     /// `true` when the aggregator successfully talked to the peer;
     /// `false` when the RPC failed. Frontend uses this to badge the
     /// card as offline.
@@ -395,6 +401,8 @@ impl NodeStatusV2 {
             cache: r.cache,
             timers: r.timers,
             daemon_started_unix: r.daemon_started_unix,
+            cpu_load_1m: r.cpu_load_1m,
+            health_score: r.health_score,
             online: true,
             error: None,
         };
@@ -421,6 +429,8 @@ impl NodeStatusV2 {
             cache: None,
             timers: Vec::new(),
             daemon_started_unix: None,
+            cpu_load_1m: None,
+            health_score: 0,
             online: false,
             error: Some(e),
         }