All of the transmute_copy() calls in simd_funcs.rs of encoding_rs are actually safe and I'd like to write them so.
That's news to me, but happy news.
For ergonomic portable SIMD, it's essential that we have convenient safe syntax for the SIMD type conversions that are zero-cost reinterpretations in the little-endian case but produce different results in the big-endian case.
Since, thanks to WebGL, big endian is not coming back, I don't care much what Rust does for SIMD in the big endian case (compute different results, inject shuffles to match little-endian results or make the conversion unavailable when compiling for a big-endian target), but I really want to have safe wrappers for the little-endian SIMD transmutes.