Feature: Measure function ignoring setup phase

[Andersama]

Hey, I saw the other thread, hope you're doing ok.

I've been fiddling with more code recently and went to benchmark things and I forgot I wrote my own benchmark header for this exact reason.

Simple example...

int main()
{
	vector<uint32_t> values;

	pcg32_random_t rng;
	rng.state = (uint64_t)&rng;
	rng.inc   = std::chrono::steady_clock::now().time_since_epoch().count();
	uint32_t l = pcg32_random_r(&rng);
	for (size_t i = 0; i < (l&0xff); i++) {
		pcg32_random_r(&rng);
	}
	
	for (size_t i = 0; i < 4096; i++) {
		values.emplace_back(pcg32_random_r(&rng));
	}

	auto bench = ankerl::nanobench::Bench();

	bench.run("sort_highnibble", [&] {
		for (size_t i = 0; i < values.size(); i++) { //we don't want to be benchmarking this loop
			values[i] = pcg32_random_r(&rng);
		}
		sort_highnibble::sort_highnibble(values.begin(), values.end(), key); //we're trying to test for this
	});

	return 0;
}

So a while ago I made a cut down version of nanobench where two lamda's are provided, where we time between the setup and operation phase. This is all I needed, but I'm in the habit now of using nanobench on instinct, and I like the pretty printing of nanobench, so I was wondering or curious if this sounds like a handy feature to add.

[l0rinc]

Seems like a duplicate of #130

[Andersama]

Sounds like they want three phases, the setup, operation (benchmarked), and a cleanup. Feels like deja vu.