BuildKite: Make post-command more robust against stuck containers.

Some BuildKite builds have been getting stuck on this step.
Interestingly, this was not on the first container that they try to stop,
implying that the problem is container-specific, rather than the Docker
daemon being unresponsive as a whole.

This change removes the dependency on `docker inspect` and adds a timeout
to the container kill step. Will it be enough? Not sure.

#codehealth

PiperOrigin-RevId: 501132910

.buildkite/hooks/post-command

@@ -71,11 +71,12 @@ if test "${BUILDKITE_COMMAND_EXIT_STATUS}" -ne "0"; then
 fi
 
 # Kill any running containers (clear state), except for "bootstrap".
-for container in $(docker ps -q); do
-  maybe_kill="$(docker inspect -f '{{if ne "/bootstrap" .Name}}true{{ end }}' "${container}")"
-  if test -n "${maybe_kill}"; then
-    docker container kill "${container}"
+for id_and_name in $(docker ps --format='{{.ID}} {{.Names}}'); do
+  if [[ "$(echo "$id_and_name" | cut -d' ' -f2-)" == 'bootstrap' ]]; then
+    continue
   fi
+  timeout --kill-after=10s --preserve-status 8s \
+      docker container kill "$(echo "$id_and_name" | cut -d' ' -f1)"
 done
 
 set -euo pipefail