Redesign cleanup system.

[T-256]

Ref: #805

It is always preferable to clean up even if a keyboardinterrupt has been raised, because the user may catch the keyboardinterrupt in his code and continue to use httpcore.ConnectionPool.

Originally posted by @karosis88 in #805 (reply in thread)

For such cases we should refer to where those objects will re-use if we don't cleanup immediately. For example:

            try:
                connection = await status.wait_for_connection(timeout=timeout)
            except (Exception, self._cancelled_exc) as exc:
                # If we timeout here, or if the task is cancelled, then make
                # sure to remove the request from the queue before bubbling
                # up the exception.
                async with self._pool_lock:
                    # Ensure only remove when task exists.
                    if status in self._requests:
                        self._requests.remove(status)
                    raise exc

This will cleanup status in self._requests immediately if an exception occurred, IMO the well based cleanup system should not block the process, we should do the cleanup when the object needs to be cleaned, re-used again, (e.g. for current example we should check where self._requests accessed then do the cleanups.)

What I mean from block the process?
Assume user hit Ctrl+C, he expects get reaction fast and don't care about cleanups, so we should not wait for cleanup that is maybe IO blocking process.
So when do the cleanups?
Answer is object needed to be cleaned, where accessed again, there we should check if object needs to be cleaned or not!