Refactor DandelionShuffleThread to run in connection scheduler

[barrystyle]

This PR refactors the dandelionshuffle thread, to run as a scheduler function which gets called once a second.

[SmartArray]

Issue validated using Instruments.

@barrystyle is 100% right about the blocking loop that causes the CPU time of that thread to spin intensely, which is definitely far too inefficient.

The current implementation only enters DandelionShuffle() (line 1798) every once in a while. I believe that the interval of 1 second - as suggested by @barrystyle - is fairly sufficient. We may also bump that number higher.

digibyte/src/net.cpp

Lines 1791 to 1798 in 3e6a19a

	void CConnman::ThreadDandelionShuffle() {
	auto current_time = GetTime<std::chrono::microseconds>();
	auto next_dandelion_shuffle = PoissonNextSend(current_time, DANDELION_SHUFFLE_INTERVAL);
	while (!interruptNet) {
	current_time = GetTime<std::chrono::microseconds>();
	if (current_time > next_dandelion_shuffle) {
	DandelionShuffle();

Also big thumbs up to use the node's scheduler instead of a thread, which, from the looks, should be fine as DandelionShuffle() has a low complexity. This is the only thing we will need to validate once more before ACKing this PR.

[barrystyle]

just removed the actual thread definition and 'on shutdown' thread cleanup

[JohnnyLawDGB]

So basically inserting breaks between CPU requests? Genius.

[SmartArray]

I just validated that the complexity of DandelionShuffle() is low enough to run in a scheduler. It took about 0-1ms per execution, which is insignificant.

Furthermore I pushed two more commits.
3a7b9fd renamed the function as it isn't a thread anymore.
48227d4 exported the scheduler interval to an external constant.

[SmartArray]

ACK

[gto90]

ACK

Great work on this performance improvement @barrystyle . One thing I would love to start working towards is some unit test coverage around the Dandelion code, but this is great for now.

[JaredTate]

ACK. Looks good. Tested & working. Thanks for doing this Barry!