perf: Warm up the model's KV cache during startup, not on the first user request

The system prompt + tool schemas prefix (~6.4k tokens) is identical on
every request and llama.cpp caches/reuses it automatically, but only
after it's been processed once. Without a warm-up, that first prefill
(50s+ on this machine, likely longer on Spaces CPU) landed on whichever
user happened to send the first message after a cold start, looking
like the app was unresponsive. Now it happens during startup instead.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

app.py

@@ -29,7 +29,12 @@ from controlai_rag.index import ControlRAGIndex
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     print("Pre-loading ControlAI Core Engine on startup...")
-    get_agent()
+    agent = get_agent()
+    print("Warming up KV cache for the system prompt + tool schemas...")
+    try:
+        agent.run("hi", verbose=False)
+    except Exception as exc:
+        print(f"Warning: warm-up generation failed (non-fatal): {exc}")
     print("ControlAI Core Engine is online and ready for traffic.")
     yield