Pass --kubeconfig to all kubectl subprocess invocations

Description

Currently, kubectl commands invoked via subprocess do not receive the --kubeconfig flag, causing them to fall back to default kubeconfig resolution ($HOME/.kube/config or KUBECONFIG env var). This creates inconsistency when KubernetesSystem.kube_config_path points to a non-default location: the Kubernetes Python client API uses the configured path (via load_kube_config()), but kubectl subprocesses may target a different cluster.

Affected Code

All kubectl invocations in src/cloudai/systems/kubernetes/kubernetes_system.py:

• kubectl exec (line ~313)
• kubectl get pods (line ~255)
• kubectl cp (line ~325)
• kubectl delete dgd (line ~488)

Proposed Solution

Add --kubeconfig flag with the resolved kube_config_path value to all kubectl subprocess commands to ensure consistency with the Python API client configuration.

Example:

kubectl_cmd = ["kubectl", "--kubeconfig", str(self.kube_config_path), "exec", "-n", self.default_namespace, ...]

Context

Identified during review of PR #793: https://github.com/NVIDIA/cloudai/pull/793#discussion_r2745665546

Requested by: @amaslenn

Read src/cloudai/systems/kubernetes/kubernetes_system.py and inspect the kubectl invocations for exec, get pods, cp, and delete dgd. Verify each subprocess command uses the resolved self.kube_config_path, and consider how existing command construction is covered before confirming all four paths use the same configuration.