benches: reduce codspeed flakiness

[DaniPopes]

Some smaller benchmarks change results on every run; this is due to either:

• using rand::rng which is not deterministic
• so few instructions that it was benchmarking loading/storing to memory

This PR fixes both, and all benchmarks should be stable; the benchmarks still contain a lot of copies but that's out of our control: CodSpeedHQ/codspeed-rust#98

[codspeed-hq]

CodSpeed Performance Report

Merging #470 will degrade performances by 15.5%

_{Comparing DaniPopes:codspeed-flaky (1a3b792) with main (767f07d)}

Summary

⚡ 72 improvements
❌ 3 regressions
✅ 150 untouched benchmarks
⁉️ 1 dropped benchmarks

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

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	add/4096	3.9 µs	3.3 µs	+18.07%
⚡	add/512	537.2 ns	440.6 ns	+21.94%
⚡	add_mod/192	719.4 ns	564.2 ns	+27.52%
⚡	add_mod/384	2.7 µs	2.3 µs	+13.84%
⚡	add_mod/4096	10.3 µs	8.6 µs	+19.01%
⚡	add_mod/64	630.3 ns	567.8 ns	+11.01%
⁉️	algo/addmul_n/0	< 1 ns	N/A	N/A
❌	algo/div/3x2/ref	245.6 ns	273.6 ns	-10.25%
❌	algo/div/reciprocal/ref	154.4 ns	182.8 ns	-15.5%
⚡	algorithms/gcd/matrix/from_u64	331.7 ns	301.4 ns	+10.05%
❌	algorithms/gcd/matrix/from_u64_prefix	578.1 ns	664.4 ns	-13%
⚡	cmp/128	91.7 ns	29.7 ns	×3.1
⚡	cmp/192	152.2 ns	29.7 ns	×5.1
⚡	cmp/256	154.4 ns	29.7 ns	×5.2
⚡	cmp/384	275.6 ns	29.7 ns	×9.3
⚡	cmp/4096	2.9 µs	2.1 µs	+41.32%
⚡	cmp/512	280 ns	29.7 ns	×9.4
⚡	cmp/64	61.1 ns	29.7 ns	×2.1
⚡	div_rem/4096/4096	5.6 µs	5.1 µs	+11.13%
⚡	div_rem/4096/64	5.7 µs	5.1 µs	+11.62%
...	...	...	...	...

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