LES: Better timeouts + Track failed requests

[rphmeier]

Closes #3984, #4037

The timeout mechanism introduced is managed by the RequestSet struct: since the flow mechanism and request IDs allow us to make multiple requests of a peer, we need to have a sensible timeout scheme which doesn't overload the peer with requests and then drop them when they fail to respond.

The RequestSet allows the peer to respond in the worst case as slowly as if requests were only dispatched one-by-one, while allowing the peer to respond to requests in a relatively arbitrary order (but assuming that earlier-dispatched requests will tend to be responded to first -- this enforces that no request is left unanswered indefinitely to mitigate some kinds of DoS attacks).

Malformatted/bad responses now correctly have their ReqId passed to handlers as part of a disconnected peer's unfulfilled set. This is accomplished by having an id_guard which must be defused in order to access the inner ReqId upon correct decoding. If not defused, the request id will be added to the peer's failed_requests set.

[coveralls]

Coverage decreased (-1.9%) to 84.649% when pulling d515ef9 on better-timeouts into 4c94878 on master.

[coveralls]

Coverage increased (+0.2%) to 84.76% when pulling e88c62d on better-timeouts into 844b1f8 on master.