See if our unicode indexing can be optimized

[uliwitness]

AIUI, if you index into a random byte in a UTF8 string (i.e. to a code unit), you can tell where the code point for this character starts, and how long it is. So we can remember the last accessed code point and the equivalent code unit and then scan backwards or forwards more efficiently. There may be other optimizations that are possible, it's worth investigating.