spec parser: GPU duration back to 120s (within ZeroGPU cap)

spec_brick.py

@@ -100,7 +100,7 @@ if _should_load():
             _state["tok"] = tok
             _state["model"] = model.to("cuda").eval()
 
-        @spaces.GPU(duration=180)   # warm path: load from pre-cached weights + generate
+        @spaces.GPU(duration=120)   # warm path: load from pre-cached weights + generate (120s is within the ZeroGPU cap)
         def _generate(text: str) -> str:
             if _state["model"] is None:
                 _load()