Make sure we don't mask LB tests failures

In `test-lb-service.sh` we do `trap "delete_resources" EXIT` to make
sure we cleanup resources on a test failure. In there, we only fetched
the `$?` after making a check for `${AUTO_CLEAN_UP}`, which itself
alters the code to 0, so function always returns success. This means
tests can never really fail.

This commit fixes it by making sure `$ERROR_CODE` is fetched at the very
beginning of the cleanup function.

tests/e2e/cloudprovider/test-lb-service.sh

@@ -20,12 +20,12 @@ LB_SUBNET_NAME=${LB_SUBNET_NAME:-"private-subnet"}
 AUTO_CLEAN_UP=${AUTO_CLEAN_UP:-"true"}
 
 function delete_resources() {
+  ERROR_CODE="$?"
+
   if [[ ${AUTO_CLEAN_UP} != "true" ]]; then
     exit ${ERROR_CODE}
   fi
 
-  ERROR_CODE="$?"
-
   printf "\n>>>>>>> Deleting k8s services\n"
   kubectl -n ${NAMESPACE} get svc -o name | xargs -r kubectl -n $NAMESPACE delete
   printf "\n>>>>>>> Deleting k8s deployments\n"