regex-syntax: Optimize SimpleCaseFolder

[Keats]

In giallo, the PHP grammar (https://github.com/shikijs/textmate-grammars-themes/blob/4cb1625ad87a07fb4674fe97b921a8f756f4694b/packages/tm-grammars/grammars/php.json) has a lot of patterns like (?i)[_a-z\x7F-\x{10FFFF}].

With master case folding, it would iterate every codepoint and be generally slow. Replacing it with an iterator that looks at the table directly is much faster. The code is also much simpler, I'm not sure if I'm missing something tbh.

Benchmark results with the added bench from that PR:

master:

test parse_full_folding   ... bench:   2,767,636.50 ns/iter (+/- 321,494.26)
test parse_huge           ... bench:       1,175.10 ns/iter (+/- 9.38)
test parse_medium1        ... bench:       4,395.45 ns/iter (+/- 43.96)
test parse_medium2        ... bench:       3,892.35 ns/iter (+/- 25.34)
test parse_medium3        ... bench:      12,665.54 ns/iter (+/- 72.62)
test parse_simple1        ... bench:       1,126.57 ns/iter (+/- 26.19)
test parse_simple2        ... bench:       2,204.90 ns/iter (+/- 29.84)
test parse_simple_folding ... bench:         762.16 ns/iter (+/- 10.08)
test parse_small1         ... bench:      12,715.30 ns/iter (+/- 116.87)

this branch:

test parse_full_folding   ... bench:      36,826.69 ns/iter (+/- 688.75)
test parse_huge           ... bench:       1,159.02 ns/iter (+/- 9.32)
test parse_medium1        ... bench:       4,360.44 ns/iter (+/- 38.80)
test parse_medium2        ... bench:       3,897.28 ns/iter (+/- 34.22)
test parse_medium3        ... bench:      12,760.71 ns/iter (+/- 370.60)
test parse_simple1        ... bench:       1,125.55 ns/iter (+/- 14.01)
test parse_simple2        ... bench:       2,095.07 ns/iter (+/- 24.99)
test parse_simple_folding ... bench:         785.47 ns/iter (+/- 7.44)
test parse_small1         ... bench:      12,747.68 ns/iter (+/- 223.23)

So about 70x faster for that pattern and no changes for simpler ones.

Kinda related: #893