feat: add ComfyUI, rebalance GPU allocation

GPU rebalance (4× V100 32GB on worker-1):
- reasoning: 2 GPU (unchanged, PP=2 for Qwen3-32B)
- ornith: 2→1 GPU (scale to 1 replica, ornith:35b only)
- comfyui: 0→1 GPU (new)
- embeddings/reranker: 0 GPU (CPU, unchanged)

qwen2.5:3b-instruct moved to CPU on talos-cp-2 (144GB RAM).
Separate Ollama deployment + 5Gi PVC, pulls model on first start.

Gateway config updated in homelab-frontend (separate commit).

Co-authored-by: poimen <[email protected]>

k8s/apps/llm-serving/kustomization.yaml

@@ -14,6 +14,7 @@ resources:
   - ornith.yaml
   - reasoning.yaml
   - reranker.yaml
+  - qwen-cpu.yaml
   - networkpolicy.yaml
 # No namespace transformer: every file sets its own, and the transformer would
 # rewrite metadata.namespace on anything cross-namespace added later.