std.ArrayList: initial capacity based on cache line size

[andrewrk]

also std.MultiArrayList

build zig with itself (ReleaseFast, no libc)

Benchmark 1 (3 runs): master/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          12.7s  ±  106ms    12.6s  … 12.8s           0 ( 0%)        0%
  peak_rss            983MB ± 25.8MB     956MB … 1.01GB          0 ( 0%)        0%
  cpu_cycles         93.7G  ± 63.0M     93.6G  … 93.7G           0 ( 0%)        0%
  instructions        197G  ± 2.72M      197G  …  197G           0 ( 0%)        0%
  cache_references   5.69G  ± 12.6M     5.67G  … 5.70G           0 ( 0%)        0%
  cache_misses        413M  ± 2.31M      411M  …  415M           0 ( 0%)        0%
  branch_misses       491M  ±  306K      491M  …  492M           0 ( 0%)        0%
Benchmark 2 (3 runs): branch/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          12.2s  ± 34.3ms    12.1s  … 12.2s           0 ( 0%)        ⚡-  4.1% ±  1.4%
  peak_rss            940MB ± 6.92MB     935MB …  948MB          0 ( 0%)          -  4.4% ±  4.4%
  cpu_cycles         90.3G  ± 67.1M     90.3G  … 90.4G           0 ( 0%)        ⚡-  3.6% ±  0.2%
  instructions        196G  ± 6.30M      196G  …  196G           0 ( 0%)          -  0.2% ±  0.0%
  cache_references   5.57G  ± 18.0M     5.55G  … 5.59G           0 ( 0%)        ⚡-  2.2% ±  0.6%
  cache_misses        405M  ± 3.40M      401M  …  408M           0 ( 0%)          -  2.1% ±  1.6%
  branch_misses       412M  ±  512K      411M  …  412M           0 ( 0%)        ⚡- 16.1% ±  0.2%

similar to #22861

[Snektron]

What do you think about also preferring capacities which are a multiple of the cache line size?

Suggested change

	new +|= new / 2 + init_capacity;
	new = std.mem.alignForward(usize, new +| new / 2 + init_capacity, init_capacity)

[andrewrk]

worth a try. that expression needs work. probably align backward would be fewer instructions and avoid potential overflow