fix: shorten the orchestrator healthcheck's retry budget

docker-compose.yml

@@ -124,7 +124,13 @@ services:
       test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8000/healthz', timeout=2)"]
       interval: 5s
       timeout: 3s
-      retries: 20
+      retries: 10
+      # Warm-up window: failures inside start_period do not consume the retry
+      # budget, so a booting orchestrator that becomes healthy within 50s is
+      # never counted against retries. A dead service trips the gate at
+      # ~100s (start_period + 10 x 5s), matching the previous retries: 20
+      # budget exactly; the win is boot tolerance, not faster dead detection.
+      start_period: 50s
 
   backend:
     build: