perf: Process merge-string sections in more even chunks

[davidlattimore]

For string-merge sections, we now process chunks of 140KB regardless of section boundaries.

[davidlattimore]

I see the following changes in benchmarks for this change (Ryzen 9955HX with 32 threads):

clang-debug: -7.2%
clang-release: no change
librustc-driver: -2.3%
wild: -3.0%

[mati865]

Funny how good the wall time is, despite everything else glowing red:

Benchmark 2 (3 runs): ./run-with ~/Projects/wild/target/dist/wild-1345 --no-fork
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          1.93s  ± 8.19ms    1.92s  … 1.93s           0 ( 0%)        ⚡-  6.9% ±  1.5%
  peak_rss           24.5GB ± 4.82MB    24.5GB … 24.5GB          0 ( 0%)          -  0.3% ±  0.1%
  cpu_cycles          233G  ±  995M      232G  …  233G           0 ( 0%)          +  0.7% ±  1.6%
  instructions        237G  ±  723M      237G  …  238G           0 ( 0%)        💩+  3.4% ±  0.5%
  cache_references   4.23G  ± 32.5M     4.20G  … 4.26G           0 ( 0%)        💩+ 10.5% ±  1.5%
  cache_misses       1.16G  ± 6.66M     1.15G  … 1.17G           0 ( 0%)        💩+  6.6% ±  1.1%
  branch_misses      1.14G  ± 1.81M     1.13G  … 1.14G           0 ( 0%)        💩+  1.6% ±  0.3%

[MatthijsKok]

Funny how good the wall time is, despite everything else glowing red:

Benchmark 2 (3 runs): ./run-with ~/Projects/wild/target/dist/wild-1345 --no-fork
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          1.93s  ± 8.19ms    1.92s  … 1.93s           0 ( 0%)        ⚡-  6.9% ±  1.5%
  peak_rss           24.5GB ± 4.82MB    24.5GB … 24.5GB          0 ( 0%)          -  0.3% ±  0.1%
  cpu_cycles          233G  ±  995M      232G  …  233G           0 ( 0%)          +  0.7% ±  1.6%
  instructions        237G  ±  723M      237G  …  238G           0 ( 0%)        💩+  3.4% ±  0.5%
  cache_references   4.23G  ± 32.5M     4.20G  … 4.26G           0 ( 0%)        💩+ 10.5% ±  1.5%
  cache_misses       1.16G  ± 6.66M     1.15G  … 1.17G           0 ( 0%)        💩+  6.6% ±  1.1%
  branch_misses      1.14G  ± 1.81M     1.13G  … 1.14G           0 ( 0%)        💩+  1.6% ±  0.3%

What benchmark program is this? @mati865

[mati865]

@MatthijsKok this is poop mentioned in https://github.com/davidlattimore/wild/blob/b9c585a4e1a6118e5c125d8cc8a8701144149d0f/BENCHMARKING.md?plain=1#L89

FYI, this was on clang with RelWithDbgInfo profile.