perf(codegen): Use char codes for simple classes

[scttcper]

This optimizes generated parser code for simple character classes.

A lot of grammars have hot loops that look like this:

identifier = [a-zA-Z_$] [a-zA-Z0-9_$]*
number = [0-9]+ ("." [0-9]+)?
whitespace = [ \t\n\r]*

Peggy currently emits a regexp test for each character consumed by those classes. This PR emits direct charCodeAt comparisons when the class is simple enough to do that exactly: no i flag, no unicode mode, and only single-code-unit ranges/characters.

Anything more complicated still uses the existing regexp path.

As one real-world benchmark, I used Sentry’s search grammar, which is the grammar we use for parsing search bar queries: https://github.com/getsentry/sentry/blob/master/static/app/components/searchSyntax/grammar.pegjs. That grammar has a lot of key/value filters, so longer searches spend plenty of time in these character-class loops.

Input	Before	After	Speedup
200 short search strings, cycling through common filter/query shapes	3.1664 ms	2.5784 ms	18.6%
one 2.7 KB query with 120 mixed free-text, filter, and grouped terms	0.7804 ms	0.6653 ms	14.7%
two copies of the long query joined with AND, about 5.4 KB	1.6284 ms	1.4060 ms	13.7%
the long query plus a parenthesized copy joined with OR, about 5.4 KB	1.6405 ms	1.3890 ms	15.3%

For that generated parser, repeated class-check code went from 177 charCodeAt calls to 76 because each check computes the character code once and reuses it.

[scttcper]

let me know if you need the benchmark or something else