icu_unicodeset_parser uses a bunch of allocating types internally to allow for

• arbitrary-length escapes (\x{61 62 63 64...}) use Vec<char>
• arbitrary-length strings ({abcd...}) use String

These can/should probably be swapped out for types with "small lives on stack, big lives on heap" semantics.

Linking PR: #3670

Discuss/decide: Priority of UnicodeSet parsing efficiency