chore: remove unnecessary black_box in criterion

[DaniPopes]

The original code only uses black_box when dropping, not inside of the measurement.

This actually matters as black_box usually spills the value to the stack, so if the benchmark routine returns a large value, the benchmark will record also an unnecessary memcpy.

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #98 will degrade performances by 24.19%

_{Comparing DaniPopes:less-criterion-black-box (b71c93b) with main (6178dd1)}

Summary

⚡ 44 improvements
❌ 15 regressions
✅ 108 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	b	2.8 µs	2.2 µs	+25.24%
⚡	b	3.4 µs	2.5 µs	+35.29%
⚡	sleep_100ms	566.1 ns	536.9 ns	+5.43%
⚡	sleep_1ms	595.3 ns	536.9 ns	+10.86%
⚡	sleep_50ms	566.1 ns	536.9 ns	+5.43%
❌	bench_array1[10]	195.8 ns	225 ns	-12.96%
⚡	bench_array1[1]	119.2 ns	90 ns	+32.41%
❌	bench_array1[4]	125.3 ns	154.4 ns	-18.88%
❌	bench_array2[10]	195.8 ns	225 ns	-12.96%
⚡	bench_array2[1]	119.2 ns	90 ns	+32.41%
❌	bench_array2[4]	125.3 ns	154.4 ns	-18.88%
⚡	init_array[1000]	11.5 µs	9.4 µs	+22.17%
❌	init_array[4]	125.3 ns	154.4 ns	-18.88%
❌	mut_borrow	866.7 ns	895.8 ns	-3.26%
❌	add	91.4 ns	120.6 ns	-24.19%
❌	recursive_memoized[HashMap<u64, u64>, 0]	805.6 ns	893.3 ns	-9.83%
❌	mul	91.4 ns	120.6 ns	-24.19%
⚡	generate_combinations[5]	8.1 µs	7.4 µs	+10.44%
⚡	generate_combinations[6]	10.7 µs	9.4 µs	+13.83%
⚡	generate_combinations[8]	17.3 µs	15.5 µs	+11.46%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

[codspeed-hq]

CodSpeed WallTime Performance Report

Merging #98 will degrade performances by 3.01%

_{Comparing DaniPopes:less-criterion-black-box (b71c93b) with main (6178dd1)}

Summary

⚡ 8 improvements
❌ 1 regressions
✅ 145 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	large_setup	14 ns	11 ns	+27.27%
⚡	from_elem_decimal[2048]	240 ns	210 ns	+14.29%
❌	n_queens_solver[5]	9.3 µs	9.6 µs	-3.01%
⚡	permutations[4]	2.4 µs	2.3 µs	+3.59%
⚡	permutations[7]	801 µs	771.8 µs	+3.79%
⚡	generate_gray_code[2]	240 ns	233 ns	+3%
⚡	sleep_100ns	57.6 µs	26 µs	×2.2
⚡	sleep_100us	157.4 µs	134 µs	+17.48%
⚡	sleep_1us	52.3 µs	42.7 µs	+22.42%

[not-matthias]

Good catch! We definitely should follow what criterion is doing (see their code). Since we're already wrapping it with black_box in the drop we don't have to do it before as well!

LGTM!