p2p/discv5: fix reg lookup, polish code, use logger

[karalabe]

This PR does a few polishes and cleanups in the p2p discoveryv5 ticketing code.

Functionality wise it fixes a (probably harmless) bug when iterating the known topics for registration. The previous code in nextRegisterLookup created a list of topics to iterate over, and then exhausted that list (recreating if it became empty), until it found the same topic it started iterating from.

This logic is a bit convoluted, because topic queuing and iteration was spanned out over multiple methods, which both messed around some internal cache. One bug was that doing a "full round" of iteration meant that the first topic will be consumed twice. This is a harmless issue in this case.

The second potentially bigger bug was that iteration order was random due to being backed by a map, so the queue constructor could make the following queues in subsequent iterations: [a, b, c, d], [d, a, b, c]. If my nextRegisterLookup invocation starts at topic d from the first listing, which is empty, the next call will again be d, causing the method to return a 40 second sleep, even though there are plenty more topics to try.

The fix in the PR is to take the leftover queue from the previous invocation, and expand it randomly with all the topics not contained within it. This makes everything simpler: a full iteration is guaranteed to go over all the topics, without duplicates, and to correctly stop when every one of them was tried.

---

Furthermore the PR swaps tickets map[Topic]topicTickets to tickets map[Topic]*topicTickets. This is needed because the code was full of s.topics[topic].buckets == nil checks, which is essence only intended to check if s.topics[topic] existed, but in the non-pointer version they always return the zero value of the struct, making the code harder to understand.

Lastly, the PR removes all the debugLog invocations in favor of log.Trace. This allows us to use the fine grained controls out logging framework has to debug discovery issues. I've also added some minor simplifications in the logs. Now we have a much more readable stream of events to trace with:

[zsfelfoldi]

LGTM