Make this extension independent from the ACK_FRAME

[mirjak]

Timestamp information can be send independent of the ACK_FRAME by defining a new frame that does not ack packets and therefore does not carry the ack range information but only one packet number and a timestamp. For congestion control timestamp information and ack ranges that are used to detect loss which triggers a congestion window reduction are independent input and therefore there is no benefit in bundling them in the same frame. Actually not bundling makes it possible to send timestamp information more often than ACK frames which can be beneficial for the congestion control.

This would also would make the extensibility model easier when dependencies on other frames are avoided.

[ianswett]

We've discussed this in the past, and just like ECN marking, it's typically better to get the timestamp information at the same time that packets are newly acknowledged.

Timestamps actually become more valuable when sending fewer ACKs, since they provide delay information that is otherwise lost.

The existing congestion controllers that use receive timestamps consume them at the same time as they are notified of newly acked packets, which is why this design was chosen.

As we discussed today, I don't think there's a real extensibility problem here. It's easy to specify how this extension works with the existing ACK frame and MP-QUIC.