Build Kubernetes/Talos chaos harness for CefasDB

[osvaldoandrade]

Parent epic: #575

Problem

The failure modes we observed require Kubernetes and node-level tests, not only local unit tests.

Acceptance

• Harness can deploy CefasDB via Helm to a test cluster.
• It injects pod kill, node shutdown, NotReady, orphan/stale process, disk pressure, and network partition.
• It collects logs, events, metrics, and cefas-manager doctor output for each scenario.
• The harness can run locally against the hack cluster and in CI-compatible mode.