Graceful shutdown of websockets

[aviramha]

Hi!
We've encountered a situation where Kubernetes terminates a pod, and Websocket connections aren't gracefully shutdown.
I debugged it a bit, and found out how Uvicorn behaves on termination - on simple HTTP - it just cancels the keep alive then awaits the server tasks.
On the websocket, it closes the transport on currently "running" websockets.
I want to suggest changing this behavior to only stop serving new connections, without shutting down the connections.
The code change would be just removing the transport.close on each websocket implementations, leaving the server.py code to await pending tasks, and canceling those if needed.

Let me know if I can send a PR to apply this change.

[euri10]

iirc I added this while working on making pytest stricter #929 and specifically the ws addition you mention I was unsure about, see this comment for further context.

so I'd be ok seeing a PR that removes that, but we just need to be sure that :

1. it doesn't pop unrelated ResourceWarning raised by pytest,
2. that the fix would not be better handled by Enable graceful shutdown when running multiple workers and sending a SIGTERM #853 or When receiving a SIGTERM supervisors should terminate their processes before joining them #1069 if this is the case (several processes) because I'm not too sure if those kind of issues are not caused more generally by our long-standing "graceful shutdown issue"