Change to an activity based timeout

[Tratcher]

Fixes #1052
Fixes #118 (comment)

The request Timeout used to cover the time until the response headers were received, which may also include sending the request body. This timeout could be problematic for large requests or gRPC client streaming. There was no timeout on the response body or on WebSockets which risked a resource leak.

Now a single activity token and timeout is used to track 'liveness', where that timeout is reset any time there's observable progress such as:

• Response headers received
• Request body data received
• Request body data sent
• Response body data received
• Response body data sent
• WebSocket data sent or received, including pings.

What doesn't count:

• HTTP/2 pings
• TCP keep-alives

This way we're not limiting the size of a request nor the direction of the traffic, so long as there continues to be any traffic.

Server request and response body data rates may still apply (though we opt out of those for gRPC & Kestrel).

(Thanks to @MihaZupan's last change, this was pretty easy 😁, with many of the changes being aesthetic.)

[MihaZupan]

Nice, I like how non-invasive this is.

A perf note: We'll want to test what impact this has when performing many reads/writes eventually.
For example, we could avoid resetting the timer each time by allowing a 1 second grace period.

[Tratcher]

I'll keep an eye on the next benchmark run to see what the impact is.

[Kahbazi]

@Tratcher breaking-change label?

[Tratcher]

@Tratcher breaking-change label?

Yes, thanks for the reminder.