If ascii characters aren't guaranteed, would this solution still be able to scale? I wonder how the overhead from scanning the input to find the domain to use f

If ascii characters aren't guaranteed, would this solution still be able to scale? I wonder how the overhead from scanning the input to find the domain to use for the bitset compares to overhead of allocating a bunch of sets. Edit: More generally, I feel like these sorts of tricks are getting less relevant over time. With the prevalence of Unicode text how many clever interview question optimizations that assume characters are just another way of saying u8 are no longer relevant?