Represent strings as sequences of bytes instead of code points

[jayconrod]

Storing strings as sequences of 32-bit code points is inefficient in terms of memory use compared with UTF-8. Most common string operations (search, split, comparing for equality) work just as well if not better on UTF-8. Operations that do require code points (collation, normalization, other locale-specific stuff) generally works fine with iteration and does not need random access.

The built-in String class should be represent strings as an array of UTF-8 bytes. Gypsum and CodeSwitch will both need to be updated.