Document long-running connections

## Feature Request

Document long-running connections. Specifically the connection between the `linkerd-proxy` and the container it's proxying for, but also other connections.

### What problem are you trying to solve?

I was trying to diagnose a problem which was partly caused by the long-running connection between `linkerd-proxy` and the main container in our pod.

We had a NodeJS container wouldn't shut down when we `exec`ed into the container killed it. Restify (the HTTP server library) apparently stops accepting incoming connections, but doesn't terminate existing connections. The health checks were still getting through! For context, we were trying to kill containers to test our alerting.

Some documentation about the connections would have sped up the diagnosis process.

### How should the problem be solved?

Something in the docs about the connections, how work and why. Is there a way to limit how long they last? Is there some other general documentation I should read about TCP connections?

### Any alternatives you've considered?

Pointers to relevant third-party documentation.

### How would users interact with this feature?

Read the docs? :smile:

Start by locating the website documentation covering linkerd-proxy, the proxied container, and TCP connections. Determine which connection behavior, lifetime or limits should be documented, then add guidance explaining how the connections work, why they persist, and where readers can find relevant references.

Written by the indexing model from the issue text.