Consider having next_notification also return the size of the next notification

[tomaka]

At the moment, here's what a process does to receive a notification:

• The process pre-allocates some buffer to receive the notification.
• We call next_notification and jump to the kernel.
• If the notification is large than the pre-allocated buffer, we return an error plus the required buffer size.
• The process then allocates a large-enough buffer and call next_notification.

If right after we want to receive the next notification, we have to do a pre-allocation again and try again.

It would be more efficient if the kernel immediately returned as well the size of the next notification that the call would have returned, in order to skip that step the next time.

[tomaka]

cc #863