Implement Graceful Shutdown / Connection Draining

This was present in the old version of Channels. The changelog says:

0.9.4 (2016-03-08)
------------------

* Worker processes now exit gracefully (finish their current processing) when
sent SIGTERM or SIGINT.

This is no longer the case.

With the new architecture in Channels 2 this ability will need to be moved to Daphne. Daphne can not stop running. It will need some kind of API to reload new code while continuing to service existing connections on the old processes.

The issue names no files, tests, or entry points. Start by reviewing Daphne's current process and connection lifecycle, then compare it with the cited Channels 0.9.4 behavior. Done would require an agreed design and implementation for graceful SIGTERM/SIGINT shutdown, connection draining, and code reload while existing connections remain served.

Written by the indexing model from the issue text.