Fix nginx/newt boot race by raising API healthcheck start_period to 120s

On a fresh install, the api container's docker-entrypoint.sh does
migrations + seed + MaxMind GeoLite2 download synchronously before
binding port 4000, which routinely exceeds the previous 75s tolerance
(start_period 30s + 3x15s retries). When the API flipped to unhealthy,
nginx and newt (which depend on api: condition: service_healthy)
aborted, leaving the tunnel offline and requiring a manual
'docker compose up -d nginx newt' second pass.

Bumping start_period to 120s gives first-boot enough slack, matches
convention (Rocket.Chat 90s, Gancio 60s), and leaves steady-state
restarts unaffected. Change applied to both dev (docker-compose.yml)
and prod (docker-compose.prod.yml) since the release tarball ships
the latter.

Bunker Admin

docker-compose.prod.yml

@@ -30,7 +30,7 @@ services:
       interval: 15s
       timeout: 5s
       retries: 3
-      start_period: 30s
+      start_period: 120s
     environment:
       - NODE_ENV=${NODE_ENV:-production}
       - PORT=4000

docker-compose.yml

@@ -29,7 +29,7 @@ services:
       interval: 15s
       timeout: 5s
       retries: 3
-      start_period: 30s
+      start_period: 120s
     environment:
       - NODE_ENV=${NODE_ENV:-development}
       - PORT=4000