fix(controller): reap stuck-registered runners after 5min instead of 30min

A runner that registers with Gitea but never picks up a job (because the
queued jobs that triggered its spawn turned out to be waiting on
'needs:' / gated by 'if:' conditions, or any other claim-side stall)
holds capacity for the full window. With a 30-minute threshold, a burst
of over-eager spawns can block all real work for half an hour.

Drop the threshold to 5 minutes. False positives are self-healing: if a
job was about to be claimed, the next brew tick (5s) will see it still
queued and spawn a fresh runner — cost is one extra image pull (cached)
and a registration token round-trip.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

crates/gongfoo-controller/src/brew.rs

@@ -682,14 +682,16 @@ async fn reap_stale(
         mark_failed(pool, *runner_id, "reaped: stuck started > 60s").await?;
     }
 
-    // --- Stuck registered > 30min (registered but never picked up a job) ---
+    // --- Stuck registered > 5min (registered but never picked up a job) ---
+    // False positives are self-healing: if a job was about to be claimed, the
+    // next brew tick will spawn a fresh runner for the still-queued job.
     let stuck_registered = sqlx::query_as::<_, (Uuid, Option<String>, String)>(
         r#"
         SELECT r.id, r.container_id, h.agent_endpoint
         FROM runners r
         JOIN hosts h ON h.id = r.host_id
         WHERE r.state = 'registered'
-          AND COALESCE(r.registered_at, r.updated_at) < now() - interval '30 minutes'
+          AND COALESCE(r.registered_at, r.updated_at) < now() - interval '5 minutes'
         "#,
     )
     .fetch_all(pool)
@@ -705,12 +707,12 @@ async fn reap_stale(
                         runner_id: *runner_id,
                         container_id: cid.clone(),
                         force: false,
-                        reason: "stuck in registered > 30min".to_owned(),
+                        reason: "stuck in registered > 5min".to_owned(),
                     },
                 )
                 .await;
         }
-        mark_failed(pool, *runner_id, "reaped: stuck registered > 30min").await?;
+        mark_failed(pool, *runner_id, "reaped: stuck registered > 5min").await?;
     }
 
     // --- Silent runner: running but no update for 10 minutes ---