fix(runtime): use unified start.py to keep FastAPI & Gradio alive on HF

Dockerfile

@@ -53,4 +53,4 @@ HEALTHCHECK --interval=30s --timeout=3s --start-period=10s --retries=3 \
 EXPOSE 7860
 
 # Run both FastAPI (backend) and Gradio (frontend)
-CMD ["sh", "-c", "uvicorn app.main:app --host 0.0.0.0 --port 8000 --proxy-headers --workers ${UVICORN_WORKERS:-1} & python -m demo.app"]
+CMD ["python", "start.py"]

start.py

@@ -0,0 +1,29 @@
+import subprocess
+import threading
+import os
+
+
+def run_fastapi():
+    subprocess.run(
+        [
+            "uvicorn",
+            "app.main:app",
+            "--host",
+            "0.0.0.0",
+            "--port",
+            "8000",
+            "--proxy-headers",
+            "--workers",
+            str(os.getenv("UVICORN_WORKERS", 1)),
+        ]
+    )
+
+
+def run_gradio():
+    pass  # noqa: just import to launch
+
+
+if __name__ == "__main__":
+    t = threading.Thread(target=run_fastapi, daemon=True)
+    t.start()
+    run_gradio()