Wrap websocket flush in timeout

[joshua-spacetime]

Description of Changes

This patch:

1. Breaks out of the websocket client actor loop on a close frame
2. Adds comments as to why this is safe to do
3. Wraps websocket flushes in a timeout, breaking out of the loop on timeout
4. Adds comments
5. Adds logging

The main change here is 3. Because if the flush stalls for any reason, before this patch, we could wait indefinitely without polling the other futures in the select! arms such as the liveness tick and the websocket receiver.

The flush could stall if the receiver stops ack'ing packets or if the tcp send buffer is full.

API and ABI breaking changes

None

Expected complexity level and risk

The change itself is not complicated, but the select! loop is. I'm fairly certain though that this change won't have any unintended consequences. It certainly doesn't poll fewer futures - it polls more.

Testing

Ran a large scale performance test and observed these websocket actors being dropped and cleaned up on disconnects (initiated from both server and client)

[cloutiertyler]

New changes also look good to me, since they're mostly mechanical in nature. Would be great to test, but I think it falls under mechanical enough.