fix: Add cleanup steps to prevent kuttl namespace deletion timeouts

[adwk67]

Summary

• Add a final cleanup test step to all 13 kuttl test suites that use (or may use)
  the KubernetesExecutor, to prevent namespace deletion timeouts.
• The cleanup deletes the AirflowCluster CR (triggering orderly StatefulSet scale-down),
  waits up to 120s for pods to terminate, then force-deletes any stragglers.
• Conditional (.yaml.j2) for tests that parameterise the executor type; unconditional
  (.yaml) for tests that always use the KubernetesExecutor.

Root cause

The Vector sidecar container in KubernetesExecutor DAG task pods runs Vector as a
background process (vector ... &), with bash as PID 1. When SIGTERM arrives during
namespace deletion:

1. Bash (PID 1) receives SIGTERM but has no trap — it waits for children.
2. Vector (not PID 1) ignores SIGTERM entirely.
3. The _STACKABLE_POST_HOOK in the base container (sleep 10; touch shutdown) fails
   because sleep itself gets killed by SIGTERM before creating the shutdown file.
4. The pod sits in Terminating for the full terminationGracePeriodSeconds (300s).

This was not visible before the kuttl v0.11.1 → v0.20.0 bump (2026-04-22), because
kuttl v0.11.1 fired namespace deletion and moved on without waiting.

Proper fix (operator-rs)

This PR is a workaround. The proper fix belongs in operator-rs
(crates/stackable-operator/src/product_logging/framework.rs, around line 1444).
There is already a commented-out alternative in the code (lines 1440–1443) that
uses exec to make Vector PID 1 so it receives and handles SIGTERM directly:

// bash -c 'sleep 1 && if [ ! -f "...shutdown" ]; then mkdir -p ... && inotifywait ...; fi && kill 1' &
// exec vector --config ...

This approach should be completed and enabled. Once that fix lands in operator-rs,
these cleanup steps can be removed.

Test plan

• Ran all 11 KubernetesExecutor test variants locally with cleanup steps
• Critical test (KE + Vector logging) namespace Terminating time dropped from ~7 min to ~54s
• CeleryExecutor variants unaffected (Jinja2 conditional renders empty file)
• No test failures introduced

--- PASS: kuttl (3524.76s)
    --- PASS: kuttl/harness (0.00s)
        --- PASS: kuttl/harness/mount-dags-configmap_airflow-latest-3.1.6_openshift-false_executor-kubernetes (318.34s)
        --- PASS: kuttl/harness/ldap_airflow-latest-3.1.6_ldap-authentication-server-verification-tls_openshift-false_executor-kubernetes (223.52s)
        --- PASS: kuttl/harness/remote-logging_airflow-latest-3.1.6_openshift-false_executor-kubernetes (339.06s)
        --- PASS: kuttl/harness/external-access_airflow-3.1.6_openshift-false_executor-kubernetes (175.36s)
        --- PASS: kuttl/harness/triggerer_airflow-latest-3.1.6_openshift-false_executor-kubernetes (251.92s)
        --- PASS: kuttl/harness/smoke_airflow-3.1.6_openshift-false_executor-kubernetes (211.90s)
        --- PASS: kuttl/harness/logging_airflow-3.1.6_openshift-false_executor-kubernetes (615.99s)
        --- PASS: kuttl/harness/ldap_airflow-latest-3.1.6_ldap-authentication-no-tls_openshift-false_executor-kubernetes (212.74s)
        --- PASS: kuttl/harness/ldap_airflow-latest-3.1.6_ldap-authentication-insecure-tls_openshift-false_executor-kubernetes (338.48s)
        --- PASS: kuttl/harness/mount-dags-gitsync_airflow-latest-3.1.6_openshift-false_executor-kubernetes_access-ssh (420.55s)
        --- PASS: kuttl/harness/mount-dags-gitsync_airflow-latest-3.1.6_openshift-false_executor-kubernetes_access-https (416.58s)
PASS

--- PASS: kuttl (433.99s)
    --- PASS: kuttl/harness (0.00s)
        --- PASS: kuttl/harness/versioning_airflow-latest-3.1.6_openshift-false (433.64s)
PASS

🤖 Generated with Claude Code

[NickLarsenNZ]

LGTM

The --force would normally mask the issue, but it's fine until the real fix is in.