feat: Dense 128K context + 2 slots, VLM second in Heavy tier

- Dense GPU_CONTEXT: 192K→128K (131072) to free VRAM
- Dense max_concurrent: 1→2 (VRAM now sufficient)
- Heavy tier: Dense → VLM → MoE (VLM handles 262K context)
- Total slots: 6 (2 Dense + 2 MoE + 2 VLM)

Distribution target: Dense 50%, VLM 30%, MoE 20%

NOTE: Requires llama.cpp restart on 192.168.68.8 with --ctx-size 131072

router/router.py

@@ -20,14 +20,14 @@ GPU_URLS = {
 # Max concurrent requests per GPU (based on llama.cpp --parallel)
 GPU_MAX_CONCURRENT = {
     "qwen3.6-35B-A3B": 2,   # 2 slots (Dense-first routing reduces thermal load)
-    "qwen3.6-27B-code": 1,  # 1 slot (24GB VRAM saturated at 256K ctx)
+    "qwen3.6-27B-code": 2,  # 2 slots (128K context frees VRAM)
     "qwen3.5-9b-vlm": 2,       # 2 slots (12GB VRAM, 4GB headroom)
 }
 
 # Context window sizes (tokens) — used for compaction signals
 GPU_CONTEXT = {
     "qwen3.6-35B-A3B": 262144,
-    "qwen3.6-27B-code": 196608,
+    "qwen3.6-27B-code": 131072,
     "qwen3.5-9b-vlm": 262144,
 }
 
@@ -233,9 +233,9 @@ def route(rd, tier):
         result = select_best_gpu(candidates, "medium")
         if result: return result
     
-    # TIER 4: Heavy reasoning — Dense first (thermal), MoE fallback (192K/262K context)
+    # TIER 4: Heavy reasoning — Dense first, VLM second (262K ctx), MoE last (thermal)
     if t > 25000 or turns > 15:
-        candidates = [m for m in ["qwen3.6-27B-code","qwen3.6-35B-A3B","qwen3.5-9b-vlm"] if m in avail]
+        candidates = [m for m in ["qwen3.6-27B-code","qwen3.5-9b-vlm","qwen3.6-35B-A3B"] if m in avail]
         result = select_best_gpu(candidates, "heavy_reasoning")
         if result: return result