Subsize 【 subscalar 】 encoding

Indexing strings by character (Unicode scalar) instead of UTF offsets can be more comfortable for parsers or short manipulations. The minimum cost for reading a >= U+FE character is: Lookup rope table by byte key (holding a high bit position), in goal of the scalar value in a sequence of higher fixed scalars Check for collision subtable (is not null => repeat the step 1), specializing the target high bit further For optimal use, contiguous sequences of characters after Latin- that turn into reserves into the mask sequence still allow for a few trailing Latin- (e.g. Katakana can still be mixed with Latin-* whitespace), so those trails turn into reseves as well.