InternPool: use sequential string indices instead of byte offsets

[jacobly0]

Building the compiler with -Ddev=cbe:

Benchmark 1 (58 runs): master/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          2.09s  ± 7.58ms    2.07s  … 2.12s           2 ( 3%)        0%
  peak_rss            609MB ± 5.82MB     589MB …  619MB          1 ( 2%)        0%
  cpu_cycles         22.0G  ± 60.0M     21.9G  … 22.1G           0 ( 0%)        0%
  instructions       47.6G  ± 5.30M     47.6G  … 47.6G           2 ( 3%)        0%
  cache_references   2.15G  ± 10.5M     2.12G  … 2.17G           0 ( 0%)        0%
  cache_misses        106M  ± 1.61M      103M  …  110M           0 ( 0%)        0%
  branch_misses       114M  ±  605K      113M  …  116M           1 ( 2%)        0%
Benchmark 2 (58 runs): 25384/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          2.09s  ± 8.01ms    2.07s  … 2.11s           0 ( 0%)          -  0.2% ±  0.1%
  peak_rss            613MB ± 6.55MB     598MB …  625MB          0 ( 0%)          +  0.6% ±  0.4%
  cpu_cycles         22.0G  ± 66.3M     21.9G  … 22.2G           0 ( 0%)          +  0.3% ±  0.1%
  instructions       47.7G  ± 5.07M     47.7G  … 47.7G           1 ( 2%)          +  0.2% ±  0.0%
  cache_references   2.12G  ± 10.2M     2.10G  … 2.14G           0 ( 0%)        ⚡-  1.2% ±  0.2%
  cache_misses        106M  ± 1.43M      102M  …  109M           0 ( 0%)          -  0.5% ±  0.5%
  branch_misses       113M  ±  555K      112M  …  115M           1 ( 2%)          -  0.2% ±  0.2%
Benchmark 3 (58 runs): 25464/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          2.09s  ± 7.19ms    2.08s  … 2.11s           1 ( 2%)          +  0.0% ±  0.1%
  peak_rss            608MB ± 6.32MB     588MB …  618MB          3 ( 5%)          -  0.2% ±  0.4%
  cpu_cycles         22.0G  ± 49.4M     21.9G  … 22.1G           1 ( 2%)          +  0.1% ±  0.1%
  instructions       47.6G  ± 3.29M     47.6G  … 47.6G           1 ( 2%)          +  0.0% ±  0.0%
  cache_references   2.12G  ± 8.34M     2.10G  … 2.15G           2 ( 3%)        ⚡-  1.3% ±  0.2%
  cache_misses        106M  ± 1.56M      102M  …  109M           0 ( 0%)          -  0.1% ±  0.5%
  branch_misses       113M  ±  636K      112M  …  114M           0 ( 0%)          -  0.5% ±  0.2%

Where #25384 is not a measurable enough speedup to be worth the extra complexity, especially for something that needs to be backported into 0.15.2.

Supercedes #25384
Closes #22867
Closes #25297
Closes #25339