heuristic to skip chunks of leading whitespace when parsing

[samyron]

This might be a stretch but this PR implements a heuristic in json_eat_whitespace . If the next character is a \n, it may be followed by consecutive spaces (0x20). If so, we can skip them pretty quickly.

activitypub-pretty.json was generated by JSON.pretty_generate(JSON.load_file('activitypub.json')).

Compared to master on my M1 Macbook Air.

== Parsing activitypub.json (58160 bytes)
ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
               after   935.000 i/100ms
Calculating -------------------------------------
               after      9.600k (± 0.5%) i/s  (104.17 μs/i) -     48.620k in   5.064774s

Comparison:
              before:     9452.1 i/s
               after:     9599.9 i/s - same-ish: difference falls within error


== Parsing activitypub-pretty.json (65761 bytes)
ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
               after     1.177k i/100ms
Calculating -------------------------------------
               after     11.724k (± 1.6%) i/s   (85.30 μs/i) -     58.850k in   5.021088s

Comparison:
              before:    11074.0 i/s
               after:    11723.6 i/s - 1.06x  faster


== Parsing twitter.json (567916 bytes)
ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
               after    99.000 i/100ms
Calculating -------------------------------------
               after    994.981 (± 0.6%) i/s    (1.01 ms/i) -      5.049k in   5.074626s

Comparison:
              before:      909.8 i/s
               after:      995.0 i/s - 1.09x  faster


== Parsing citm_catalog.json (1727030 bytes)
ruby 3.4.2 (2025-02-15 revision d2930f8e7a) +YJIT +PRISM [arm64-darwin24]
Warming up --------------------------------------
               after    55.000 i/100ms
Calculating -------------------------------------
               after    549.774 (± 0.7%) i/s    (1.82 ms/i) -      2.750k in   5.002283s

Comparison:
              before:      435.4 i/s
               after:      549.8 i/s - 1.26x  faster

Looking for 8 spaces was slightly faster than looking for 4. I'm not sure this is safe but it might be slightly faster:

if (chunk != 0x2020202020202020) {
    if (((uint32_t) chunk) == 0x20202020) {
        state->cursor += 4;
    }
    break;
}

[byroot]

I think we can do even better.

Unless I'm mistaken, we can get the exact number of consecutive spaces with:

__builtin_ctzll(bytes ^ 0x2020202020202020) / 8

[byroot]

Improved version: #886