ts_tunnel: session rotation handshake never gives up

[danderson]

Section 6.4 of the spec specifies that rekey handshakes should give up after REKEY_ATTEMPT_TIME handshake attempts without an answer. We currently retry indefinitely.

[danderson]

This is a good opportunity to clean up the code around handshake management, which is currently splatted across endpoint.rs and handshake.rs with poor encapsulation. This makes it easy to introduce bugs because the flow of logic is hard to follow.

After staring at it, I think I want to fix this by doing that refactor now, so that I can then confidently say that we're handling handshake timeouts correctly. That means three steps, I think:

• Rework IdMap to hand out handles for session IDs it creates, so that session ID cleanup is a simple RAII style cleanup when the relevant handshake or session is dropped. That would also have avoided ts_tunnel: session IDs not deleted from IdMap when receiving back-to-back handshakes #333.
• Rework Handshake to move a bunch of the state management logic that currently lives in endpoint.rs into the Handshake enum, so that endpoint.rs can just tell the handshake what thing it wants to execute and let the handshake itself decide how to transition between states.
• Finally, have the handshake track the number of retransmit attempts, and give up per 6.4 in the spec.