ipykernel disallows jupyter user to create/attach to event loop

Hello,

as follow-up on https://github.com/jupyter/notebook/issues/3397#issuecomment-419474214, here a minimal example that I expect to finish successfully when sent to the kernel by jupyter:

```py
import asyncio
print(asyncio.get_event_loop().run_until_complete(asyncio.sleep(0)))
```

It runs fine in latest IPython which doesn't use tornado under the hood.

However, `notebook>=5.7.9` requires `tornado>=5` and so when ipykernel executes `loop = tornado.ioloop.IOLoop.current()`, an asyncio event loop will be created and `run_forever` by ipykernel.

This consequently leaves the user without options (left aside `nest-asyncio` which works for most cases), as there is a running asyncio loop on startup.

If IPython can run this simple snippet, shouldn't jupyter be able to do the same? Currently, any package that does a plain loop setup and teardown will crash in jupyter envs that have `tornado>=5`.

Reproduce the issue in a Jupyter environment using notebook>=5.7.9 and tornado>=5, starting with the asyncio snippet in the report. Inspect ipykernel's use of tornado.ioloop.IOLoop.current() and its run_forever behavior. Done means the snippet can create, run, and tear down its event loop successfully when sent to the kernel.

由索引模型根据 Issue 内容生成。