Refactors how Slic keeps connections alive

[bernardnormier]

This PR refactors how Slic keeps connections alive.

Fixes #3665.

With this new implementation, Slic keeps connection alive by sending Ping frames from client connections (like before).

The algorithm is now:

• send a Ping frame idle timeout * 0.6 after the last write even if there is an outstanding pong (= like before, except it's now 0.6 instead of 0.5)
• send a Ping frame idle timeout * 0.5 after the last read unless there is already an outstanding pong (new)

This PR also refactors the idle timeout decorator and associated tests. They are now 2 separate decorators, one for Ice and one for Slic. This PR didn't change the Ice logic.

[externl]

Looks good. See comments.

[InsertCreativityHere]

Looks sound to me!

[pepone]

Would be better to keep IDuplexConnection and then do

if (_duplexConnection is SlicDuplexConnectionDecorator duplexConnectionDecoator)
{
    duplexConnectionDecoator.Enable();
} 

Then we can make SlicDuplexConnectionDecorator timers non-nullable, and keep a single constructor.

[bernardnormier]

I don't see how this would work.

The difficulty here is the idle timeout is negotiated during connection establishment. At construction time, we don't know if the negotiated idle timeout is good to be infinite or some other value.

[pepone]

We can pass the idleTimeout I forget to add it there, and that is not the point. The point is to avoid creating this decorator for the Server connections.

[bernardnormier]

We need the decorator for server connections too. It implements the idle timer.

[pepone]

What is ReadKeepAlive there is no such operation?

[bernardnormier]

Fixed. It's the current local function, SendReadPing.

[pepone]

is outstanding Pong. correct here?

[bernardnormier]

Replaced by pending Pong, like the field name.