feat(ollama): add persistence in Ollama container

Re-enable qwen3.5:9b and qwen3.5:9bctxSmall using fresh unsloth/Qwen3.5-9B-GGUF
quantization, which uses the correct rope.dimension_sections format (4 elements)
compatible with this llama.cpp build. Both models include the mmproj for
multimodal support. The old Ollama-extracted GGUF (mrope_sections, 3 elements)
has been removed.

ollama/llama-swap.yaml

@@ -2,10 +2,27 @@ healthCheckTimeout: 180
 logLevel: info
 
 models:
-  # NOTE: qwen3.5:9b and qwen3.5:9bctxSmall are disabled — the GGUF extracted
+  "qwen3.5:9b":
-  # from Ollama uses the old mrope_sections format (3 elements) which this
+    cmd: |
-  # llama.cpp build rejects. Download a fresh quantization from HuggingFace
+      /app/llama-server
-  # (e.g. bartowski/Qwen3.5-9B-GGUF) and add them back.
+      --host 0.0.0.0 --port ${PORT}
+      --model /models/Qwen3.5-9B-Q4_K_M.gguf
+      --mmproj /models/Qwen3.5-9B-mmproj-F16.gguf
+      --alias qwen3.5:9b
+      --n-gpu-layers 999
+      --ctx-size 8192
+      --temp 1 --top-k 20 --top-p 0.95 --presence-penalty 1.5
+
+  "qwen3.5:9bctxSmall":
+    cmd: |
+      /app/llama-server
+      --host 0.0.0.0 --port ${PORT}
+      --model /models/Qwen3.5-9B-Q4_K_M.gguf
+      --mmproj /models/Qwen3.5-9B-mmproj-F16.gguf
+      --alias qwen3.5:9bctxSmall
+      --n-gpu-layers 999
+      --ctx-size 131072
+      --temp 1 --top-k 20 --top-p 0.95 --presence-penalty 1.5
 
   "hf.co/HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive:q4_k_m":
     cmd: |