fix: set tmpfs to 16G, add dockerd startup timeout with error log dump

- Change tmpfs from size=0 (default, could be too small on Modal) to 16G
- Replace infinite until loop with 60-second timeout that dumps
  /var/log/dockerd.log on failure so we can diagnose startup issues
- Sandbox now exits cleanly instead of spinning forever if dockerd fails

Author: manascb1344

app.py

@@ -176,7 +176,7 @@ def _get_gpu_config(gpu_key: str):
 # gVisor requires tmpfs for Docker overlay2 storage driver.
 # Without this, Docker falls back to the vfs driver (10-50x slower).
 mkdir -p /var/lib/docker
-mount -t tmpfs -o size=0 tmpfs /var/lib/docker
+mount -t tmpfs -o size=16G tmpfs /var/lib/docker
 
 exec /usr/bin/dockerd --iptables=false --ip6tables=false > /var/log/dockerd.log 2>&1
 """
@@ -740,7 +740,8 @@ async def github_webhook(request: Request):
         cmd = (
             "bash -c '/start-dockerd.sh &'"
             " && sleep 5"
-            " && until docker info > /dev/null 2>&1; do sleep 1; done"
+            " && for i in $(seq 1 60); do docker info > /dev/null 2>&1 && break; sleep 1; done"
+            " && docker info > /dev/null 2>&1 || { echo \"dockerd failed to start\"; cat /var/log/dockerd.log 2>/dev/null; exit 1; }"
             " && (docker load -i /images/node-22-alpine.tar || true)"
             " && cd /actions-runner"
             " && export RUNNER_ALLOW_RUNASROOT=1"