fix(ci): expand DinD loopback for tracer-release system tests (#3871)

* fix(ci): expand DinD loopback volume for system tests

System Tests jobs running on docker-in-docker µVM runners hit "no space
left on device" with the default 20G loopback (especially the parametric
scenario, which builds many container images). Set the gbi-dind env vars
to grow the volume to 50G and back it on the kubefs root device instead
of the tmpfs-backed /tmp, freeing the volume from the pod's memory budget.

See https://datadoghq.atlassian.net/wiki/spaces/K8S/pages/2874901299/How+to+use+Micro+VMs#DinD-in-CI

* fix(ci): scope DinD loopback override to tracer-release only

The previous commit applied DOCKER_LOOPBACK_SIZE=50G + DOCKER_LOOPBACK_PATH=/
to the shared .system_tests template, which broke every system-tests job:
mkfs.ext4 on the µVM root FS (virtiofs) takes longer than the runner's
hardcoded 60s wait-for-docker:2375 timeout, so dockerd never becomes
reachable. Move the size bump to tracer-release only (the original failing
job) and drop the PATH override — /tmp on these runners is real disk, not
tmpfs, so a 50G loopback there works without changing the daemon-ready
timing.

Author: Leiyks

.gitlab/generate-package.php

@@ -1321,6 +1321,10 @@
 "System Tests: [tracer-release]":
   extends: .system_tests
   timeout: 4h
+  variables:
+    # Expand the DinD loopback volume to avoid running out of disk space.
+    # See https://datadoghq.atlassian.net/wiki/spaces/K8S/pages/2874901299/How+to+use+Micro+VMs#DinD-in-CI
+    DOCKER_LOOPBACK_SIZE: 50G
   rules:
     - if: $CI_COMMIT_REF_NAME == "master"
       when: on_success