A `char` is a `u32`, but Unicode is standardized to only take up 21bit, so there is more than enough space for that optimization. Might be useful for specifying code page conversion tables with holes in them.