Problem

The llmkube cache list command currently lists Model CRD resources and their cache keys, not the actual contents of the model cache PVC.

This means:

• If Model CRs are deleted but cached models remain in the PVC, they won't be shown
• Users can't see what's actually consuming storage in the cache PVC
• No way to discover "orphaned" cache entries

Current Behavior

llmkube cache list -n my-namespace
# Only shows Model CRs that have cacheKey in status
# Returns "No models found" if no Model CRs exist, even if PVC has cached models

Expected Behavior

The command should show actual cached models by:

1. Finding a pod that mounts the llmkube-model-cache PVC in the namespace
2. Exec'ing into the pod to list /models/ directory contents
3. Cross-referencing with Model CRs to show which are "active" vs "orphaned"

Example output:

Model Cache Entries (namespace: my-namespace)
═══════════════════════════════════════════════════════════════════════════════
CACHE KEY         SIZE      STATUS      MODELS              SOURCE
a3b8c9d4e5f67890  4.1 GiB   active      llama-2-7b          ...TheBloke/Llama-2-7B-GGUF/...
f1c314277254a2fd  7.2 GiB   orphaned    -                   (unknown)

Total: 2 cache entries (1 active, 1 orphaned), 11.3 GiB used

Additional Improvements

• Show disk usage from actual PVC
• Add --orphaned flag to show only orphaned entries
• Add ability to clean orphaned entries: llmkube cache clear --orphaned
• Better feedback: "Checking namespace X..." when running