This is a little tricky, because we have to map the utf8 indices
returned by the regex engine to utf16 indices usable by Ruffle.
To limit the impact on performance, the regex, the string we're
currently matching on, and the last known (utf8, utf16) positions
are cached, avoiding extra utf8 conversions in common use cases
where a single string is repeatedly searched with increasing
`lastIndex`.
This generally means that methods are more efficient, as we
don't need to encode to UTF16 on-the-fly to have correct indices.
This also fix some bugs:
- charCode now properly handle surrogate pairs
- calling lastIndexOf with the empty pattern and an OoB index now
properly returns the string length
Still missing is AVM2's String.match
These types represent an UCS2 string (UTF-16 with unpaired surrogates).
The string is stored either as a sequence of u8s (Bytes) or u16s (Wide);
the type of string is tracked by setting the high bit of the length of
its fat pointer.
Moulins