feat(ai): align primary model with mlx-proxy actually loaded model

mlx-proxy on the mac mini currently loads
mlx-community/gemma-4-26b-a4b-it-8bit, but config.yaml was still
requesting mlx-community/Qwen3.5-35B-A3B-4bit. The proxy was silently
serving the loaded model regardless, but the mismatch made debugging
and log tracing harder.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

config.yaml

@@ -6,8 +6,8 @@ ai:
 
   models:
     primary:
-      endpoint: "http://100.76.254.116:8800/v1/chat/completions"
-      model: "mlx-community/Qwen3.5-35B-A3B-4bit"
+      endpoint: "http://100.76.254.116:8801/v1/chat/completions"
+      model: "mlx-community/gemma-4-26b-a4b-it-8bit"
       max_tokens: 4096
       timeout: 60