Improve parsing performance

[blakeembrey]

Almost a 100% performance improvement with these parsing changes, notably removing .split() and removing .trim(), and using a loop instead of a regex to validate numbers. This is the rough order of optimization and improvement:

Before:

 ✓ src/index.bench.ts > parseRange 2205ms
     name                            hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · 1000 byte range       4,993,061.92  0.0001  1.9072  0.0002  0.0002  0.0003  0.0003  0.0010  ±0.87%  2496531
   · multiple byte ranges  1,361,804.35  0.0006  0.2075  0.0007  0.0007  0.0009  0.0010  0.0037  ±0.43%   680903

Removing array assignment (use object return):

 ✓ src/index.bench.ts > parseRange 2240ms
     name                            hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · 1000 byte range       5,323,845.39  0.0001  1.6439  0.0002  0.0002  0.0002  0.0003  0.0009  ±0.93%  2661923
   · multiple byte ranges  1,394,318.85  0.0006  0.2097  0.0007  0.0007  0.0009  0.0010  0.0037  ±0.49%   697160

Validate number in loop and remove .trim():

 ✓ src/index.bench.ts > parseRange 2448ms
     name                            hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · 1000 byte range       6,607,117.15  0.0000  1.8297  0.0002  0.0002  0.0002  0.0002  0.0003  ±1.00%  3303559
   · multiple byte ranges  1,731,118.69  0.0005  0.2345  0.0006  0.0006  0.0007  0.0007  0.0031  ±0.47%   865560

Remove .split() and use string slicing:

 ✓ src/index.bench.ts > parseRange 3023ms
     name                            hz     min     max    mean     p75     p99    p995    p999     rme  samples
   · 1000 byte range       9,609,063.96  0.0000  1.9986  0.0001  0.0001  0.0001  0.0002  0.0004  ±1.17%  4804533
   · multiple byte ranges  2,480,507.17  0.0003  0.2010  0.0004  0.0004  0.0005  0.0005  0.0010  ±0.32%  1240254

The parser is fairly similar to cookie now, so I knew the optimizations to make already. I think there's still something in the handling of missing values but it seems negligible for the effort.

[blakeembrey]

Part of the change is removing options.combineRanges in favor of a separate utility. The change is negligible for performance so it could remain an option 🤷

It is a code path that never needs checking if it's included in downstream users code though, and it makes it easier to perf test, which is why I split it apart.