Add details on how spurious losses may be detected

[ianswett]

Fixes #3058

[martinthomson]

This doesn't seem particularly actionable. I mean, I could probably work something out, but if the goal is to provide a spec where it is relatively easy to follow the advice and produce a good implementation, then this doesn't meet that test.

[Ralith]

I read this as "On acknowledgement of a packet, discard state for packets that were declared lost prior to the acknowledged packet's sending," which is at least enough to give me some ideas.

[janaiyengar]

I agree with @martinthomson. I don't want to ask for a detailed mechanism for doing this, but without it, the text here is not be particularly useful. Let's close this out.