As discussed in #372, in order to make the JuliaSyntax parsing independent of the version of Julia in which it runs, it needs to:

1. Add a dependency on utf8proc_jll, rather than using the utf8proc bundled with Julia. This is important because the supported version of Unicode (for character classification) depends on utf8proc.
2. Re-implement Base.is_id_start_char and Base.is_id_char to identify which characters can appear in identifiers (or technically, which characters can begin graphemes, now that handle ZWJ and emoji sequences, don't break identifiers within graphemes #372 is merged … this could simplify the rules slightly since we no longer need to explicitly check for combining marks).