perf: preload all checkpoints at boot so routing never pays a model load

app.py

@@ -52,6 +52,12 @@ try:
 except Exception as e:
     print("warmup skipped:", type(e).__name__, e, flush=True)
 
+if ROUTING_OK:                 # build and warm every routed checkpoint before the UI opens
+    try:
+        print(R.warmup(), flush=True)
+    except Exception as e:
+        print("routing warmup skipped:", type(e).__name__, e, flush=True)
+
 MODEL_REPO = D.MODEL_REPO
 ANSWER_COLS = ["question", "answer", "confidence"]
 
@@ -116,7 +122,7 @@ def run_playground(state_text, questions_text):
 def run_routing(text):
     """Routing tab. Wrapped so a failure shows in the tab instead of breaking the app."""
     if not ROUTING_OK:
-        return [], "**Routing unavailable** — this Space needs `laya>=0.2.1`.", "{}"
+        return [], "**Routing unavailable** — this Space needs `laya>=0.3.0`.", "{}"
     try:
         return gpu_call(R.analyse_and_answer, text)
     except Exception as e:
@@ -238,7 +244,7 @@ with gr.Blocks(title="Laya", theme=gr.themes.Soft()) as demo:
             "Khmer scores **0.000 accuracy at 0.952 confidence** on the English checkpoint."
         )
         if not ROUTING_OK:
-            gr.Markdown("> **Unavailable** — this Space needs `laya>=0.2.1`. The other tabs are unaffected.")
+            gr.Markdown("> **Unavailable** — this Space needs `laya>=0.3.0`. The other tabs are unaffected.")
         with gr.Row():
             with gr.Column():
                 ml_in = gr.Textbox(