Update docs on node draining

[andreimatei]

Re: Stop a Node

Our nodes do a couple of dances when they're being shut down so that impact on running queries, and on clients continuously sending queries, is minimized. The behavior and the timeouts involved should be documented. I think @asubiotto has become the expert.

[jseldess]

@andreimatei, I think this is documented at a high-level under "how it works". Definitely interesting in making this more detailed, though.

[asubiotto]

We could enhance that with an explanation of what happens to client connections and in-flight queries (some work being done on the distsql side of things: cockroachdb/cockroach#20450)

[jseldess]

Also involves documenting the new server.drain_max_wait cluster setting, which sets the max amount of time to wait for distributed queries to finish before the node starts draining.