Translate bytes from an mmap’d byte slice — zero syscall reads.
Uses SIMD AVX2 for range-delta patterns (e.g., a-z → A-Z).
For large data: parallel translation with rayon + single write_all syscall.
For small data: sequential chunked processing.
Translate + squeeze from mmap’d byte slice.
For large data: parallel translate with rayon, then sequential squeeze + write.
For small data: sequential translate + squeeze in-place.