[fix] Use -p/--namespace when kubeconfig context has no namespace set

[petecheslock]

Added logic to use a CLI-provided namespace if the namespace cannot be detected from kubeconfig.

User reported issue from the community slack

When a kubeconfig context has no namespace field set, the script fails with:

❌ Unable automatically detect namespace. Environment variable "LLMDBENCH_CONTROL_CLUSTER_NAMESPACE". 
Specify namespace via CLI option "-p/--namespace" or environment variable "LLMDBENCH_HARNESS_NAMESPACE"

This occurs even when -p/--namespace is provided, because:

1. LLMDBENCH_CONTROL_CLUSTER_NAMESPACE is read from the kubeconfig context (line 518)
2. If the context has no namespace, it remains empty
3. The -p flag sets LLMDBENCH_VLLM_COMMON_NAMESPACE but not LLMDBENCH_CONTROL_CLUSTER_NAMESPACE
4. The validation check at line 202 in run.sh fails because LLMDBENCH_CONTROL_CLUSTER_NAMESPACE is empty

In this scenario the user is working in the default namespace (or any namespace) whose kubeconfig context doesn't specify a namespace and they should be able to use -p/--namespace to provide it.

This solution adds a fallback after kubeconfig detection: if LLMDBENCH_CONTROL_CLUSTER_NAMESPACE is empty and LLMDBENCH_VLLM_COMMON_NAMESPACE is set (via -p), use that value.

This ensures:

• Existing behavior is preserved (kubeconfig namespace takes precedence when present)
• The -p flag works as expected when kubeconfig detection fails
• No breaking changes (only applies when detection returns empty)