on every block at least 2 threads are created and destroyed via merkle async

[spoonincode]

This std::async(),

spring/libraries/chain/include/eosio/chain/merkle.hpp

Lines 35 to 48 in 909603c

	auto async_calculate_merkle_pow2 = [&start, &size](auto fut) {
	size_t slice_size = size / fut.size();
	for (size_t i=0; i<fut.size(); ++i)
	fut[i] = std::async(std::launch::async, calculate_merkle_pow2<It>,
	start + slice_size * i, start + slice_size * (i+1));
	std::array<digest_type, fut.size()> res;
	for (size_t i=0; i<fut.size(); ++i)
	res[i] = fut[i].get();
	return calculate_merkle_pow2(res.begin(), res.end());
	};

leads to the full creation and destruction of at least 2 threads for every block. Would be nice to figure out how to use a thread pool here.

[greg7mdp]

This merkle computation is about 25% faster than the one in 5.0, even when not using std::async. And in my tests, the async version is faster still, even with the overhead of starting a thread in std::async.
I feel we should not add code complexity unless we can show it makes a significant difference. There is a perf. test for the merkle functions in benchmarks.

[bhazzard]

Not a high priority, to be explored opportunistically, considering impacts to code clarity and performance on replay / sync use cases.